Model library construction method, device, equipment, medium and product

Abstract

The application discloses a model library construction method, device, equipment, medium and product. The method comprises the following steps: preprocessing a cut section image of a laser cutting workpiece collected, and extracting a region of interest; performing image feature extraction and quantization on the region of interest, and obtaining an image feature parameter; establishing a first quantitative correlation between the image feature parameter and a multi-physical field parameter, and obtaining a first mapping model; establishing a second quantitative correlation between the multi-physical field parameter and a laser processing process parameter, and obtaining a second mapping model; and constructing a quality classification feature model library based on the first mapping model and the second mapping model. The application can improve the quality detection efficiency and effect.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method, apparatus, device, medium, and product for constructing a model library. Background Technology

[0002] With the rapid advancement of industrial automation and intelligent manufacturing, laser cutting production lines are developing towards high speed, continuous operation, and unmanned operation. This requires that the detection and judgment of cutting quality must be completed in real time, accurately, and quantitatively within the production cycle, and that the detection results can be quickly converted into adjustment schemes for process parameters in order to achieve closed-loop control and continuous optimization of the processing process.

[0003] Currently, the detection and judgment of the cross-sectional quality of laser-cut workpieces are mainly based on traditional sensor quality detection methods, such as acoustic-optical sensors and photoelectric sensors, to monitor the cutting process.

[0004] However, the above-mentioned solutions mostly develop detection algorithms for single defect types (such as only slag or only overheating), and have not yet formed a systematic model library that covers multiple dimensions of quality characteristics such as slag, overheating, roughness, and texture trends, and integrates image features and physical mechanisms. This makes it difficult to establish a quantitative correlation between the detection results and specific process adjustment schemes, and it is impossible to achieve closed-loop control from quality detection to process optimization. Summary of the Invention

[0005] The main purpose of this application is to provide a method, apparatus, device, medium and product for constructing a model library, which improves the accuracy and effectiveness of detection.

[0006] To achieve the above objectives, firstly, this application provides a method for constructing a model library, including: The acquired images of the cut surface of the laser-cut workpiece are preprocessed to extract the region of interest; Image feature parameters are obtained by extracting and quantizing the region of interest. Establish the first quantization correlation between image feature parameters and multiphysics parameters to obtain the first mapping model; A second quantitative correlation between multiphysics parameters and laser processing parameters is established to obtain a second mapping model; A quality classification feature model library is constructed based on the first mapping model and the second mapping model.

[0007] In one embodiment, the acquired cross-sectional image of the laser-cut workpiece is preprocessed to extract the region of interest, including: Adaptive filtering is applied to the cross-sectional image of the incision to obtain the filtered image; The filtered image is then subjected to histogram equalization to obtain the enhanced image. Edge detection is performed on the enhanced image to segment the cut surface region and the slag edge region, thus obtaining the region of interest.

[0008] In one embodiment, image feature extraction and quantization are performed on the region of interest to obtain image feature parameters, including: Edge contour tracking is performed on the region of interest to extract the projected area ratio, adhesion height and spatial frequency of the slag, and to distinguish between continuous and discrete slag to obtain slag feature parameters. Extract the width trend of the heat-affected zone in the region of interest and quantify the degree of overheating to obtain overheating characteristic parameters; Periodic frequency analysis of the region of interest is performed using the gray-level co-occurrence matrix to obtain roughness characteristic parameters; The angle of inclination of the cutting stripes in the region of interest relative to the normal direction of the board is measured, and the trend curve equations of each layer of texture are constructed by fitting the least squares method to obtain the texture trend characteristic parameters. Image feature parameters are composed of slag-coating feature parameters, overheating feature parameters, roughness feature parameters, and texture trend feature parameters.

[0009] In one embodiment, establishing a first quantization correlation between image feature parameters and multiphysics parameters to obtain a first mapping model includes: Energy density correlation analysis was performed on the overburning characteristic parameters and texture trend characteristic parameters to establish the first sub-mapping relationship between the actual energy density absorbed by the material per unit volume and the overburning characteristic parameters and texture trend characteristic parameters, thus obtaining the energy density mapping sub-model. A correlation analysis of airflow stripping intensity is performed on the texture trend feature parameters to establish a second sub-mapping relationship between airflow stripping intensity, boundary layering position and texture trend feature parameters, thus obtaining the airflow intensity mapping sub-model. Multiphase flow conversion efficiency is evaluated based on energy density mapping sub-model and airflow intensity mapping sub-model. A third sub-mapping relationship is established between multiphase flow conversion efficiency and actual energy density and airflow stripping intensity, resulting in a multiphase flow conversion efficiency mapping sub-model. The first mapping model is constructed by the energy density mapping sub-model, the airflow intensity mapping sub-model, and the multiphase flow conversion efficiency mapping sub-model.

[0010] In one embodiment, a second quantitative correlation is established between multiphysics parameters and laser processing parameters to obtain a second mapping model, including: The quantified values ​​of slag adhesion characteristics are fitted and correlated with laser power, processing speed and gas flow rate to construct a slag adhesion correlation sub-model; Energy density correlation analysis was performed on the overburning characteristic parameters with laser power and spot radius to construct an overburning correlation sub-model; A roughness feature parameter interval judgment standard is defined and associated with the measured texture by line-scan laser to construct a roughness association sub-model. Least square curve fitting is performed on the texture trend feature parameters and the tangent is extracted to determine the correlation between processing speed and gas flow rate, and a texture correlation sub-model is constructed. A second mapping model is constructed from the slag-attached correlation sub-model, the overheating correlation sub-model, the roughness correlation sub-model, and the texture correlation sub-model.

[0011] In one embodiment, a quality classification feature model library is constructed based on a first mapping model and a second mapping model, including: Offline machine learning training was performed on the first and second mapping models to obtain a theoretical data model benchmark library. The image feature parameters are correlated and matched with the theoretical data model benchmark library to obtain the correlation results; Based on the correlation results, the first mapping model and the second mapping model are fused and optimized to obtain a quality classification feature model library.

[0012] In one embodiment, after fusing and optimizing the first mapping model and the second mapping model based on the association results to obtain the quality classification feature model library, the method further includes: The feature parameters of the acquired laser-cut workpiece image are matched with the quality classification feature model library to obtain the quality classification result; Based on the quality classification results, the laser processing parameters in the second mapping model are matched and adjusted to obtain a recommended scheme for process path evaluation.

[0013] Secondly, embodiments of this application provide a model library construction apparatus, including: The preprocessing module is used to preprocess the acquired cross-sectional images of the laser-cut workpiece and extract the region of interest. The feature extraction module is used to extract and quantize image features from the region of interest to obtain image feature parameters; The first mapping module is used to establish the first quantization correlation between image feature parameters and multiphysics parameters to obtain the first mapping model. The second mapping module is used to establish a second quantitative correlation between multi-physics parameters and laser processing parameters, and to obtain a second mapping model. The model library building module is used to build a quality classification feature model library based on the first mapping model and the second mapping model.

[0014] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0017] This application provides a method, apparatus, device, medium, and product for constructing a model library. The method includes: preprocessing an acquired image of the cut surface of a laser-cut workpiece to extract a region of interest; extracting and quantizing image features from the region of interest to obtain image feature parameters; establishing a first quantitative correlation between the image feature parameters and multi-physics parameters to obtain a first mapping model; establishing a second quantitative correlation between the multi-physics parameters and laser processing parameters to obtain a second mapping model; and constructing a quality classification feature model library based on the first and second mapping models. This application establishes a first quantitative correlation between image feature parameters and multi-physics parameters, and a second quantitative correlation between multi-physics parameters and process parameters such as laser power, processing speed, and auxiliary gas flow rate, forming a complete technical link from image features to physical field state to process parameter adjustment. This enables quality inspection results to directly drive quantitative adjustment of process parameters, thereby improving the efficiency and effectiveness of quality inspection. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 This is a flowchart illustrating a model library construction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a model library construction device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the computer device provided in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein.

[0021] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0022] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0023] It should be understood that in this application, "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product or device.

[0024] It should be understood that in this application, "multiple" refers to two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, "and / or B" can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "Contains A, B, and C", "Contains A, B, and C" means that all three A, B, and C are contained; "Contains A, B, or C" means that one of A, B, and C is contained; "Contains A, B, and / or C" means that any one, two, or three of A, B, and C are contained.

[0025] It should be understood that in this application, "B corresponding to A", "B corresponding to A", "A corresponds to B", or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information. Matching A and B is defined as a similarity between A and B that is greater than or equal to a preset threshold.

[0026] Depending on the context, "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection."

[0027] The data involved in this application may be data authorized by the tester or fully authorized by all parties. The collection, dissemination, and use of the data shall comply with the relevant laws, regulations and standards of the relevant countries and regions. The implementation methods / executives of this application may be combined with each other.

[0028] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0029] The technical solutions of this application will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0030] Laser cutting technology, with its advantages of high precision, high efficiency, and high flexibility, has become a core process for metal sheet processing in modern manufacturing, widely used in numerous industrial fields such as automobile manufacturing, aerospace, rail transportation, and electronics. The quality of the laser-cut workpiece cross-section—including the degree of slag adhesion, overheating damage, surface roughness, and cutting stripe texture—directly determines the feasibility of subsequent processing, assembly accuracy, and the service performance and lifespan of the final product. With the rapid advancement of industrial automation and intelligent manufacturing, laser cutting production lines are developing towards high speed, continuous operation, and unmanned operation. This requires that the detection and judgment of cutting quality must be completed in real-time, accurately, and quantitatively within the production cycle, and that the detection results can be quickly converted into process parameter adjustment schemes to achieve closed-loop control and continuous optimization of the processing process. Therefore, automatic detection and process diagnosis technology for laser-cut cross-section quality based on image analysis has become an important technological foundation for realizing automatic adjustment of laser processing technology and process diagnosis of manufacturing stages.

[0031] Currently, the detection and assessment of the cross-sectional quality of laser-cut workpieces mainly rely on the following three types of technical methods: The first category is the traditional manual visual inspection and sampling method. After the product is processed, quality inspectors with relevant professional knowledge evaluate the quality of the cut surface by visual observation or with the help of simple measuring tools, or they use sampling to inspect a portion of the products. This method relies on human subjective experience and judgment, and the test results vary from person to person, lacking unified quantitative standards and reproducibility.

[0032] The second category is quality inspection methods based on traditional sensors, which typically use acoustic-optical sensors, photoelectric sensors, etc., to monitor the cutting process. These methods indirectly infer the cutting quality by collecting changes in sound, spectrum, or light signals during the cutting process. However, due to limitations in the accuracy and response characteristics of the sensors themselves, they generally suffer from insufficient detection accuracy, difficulty in achieving quantitative measurement, and high equipment costs.

[0033] The third category is image detection methods based on machine vision. This method acquires images of the cutting surface or the cutting process using industrial cameras or coaxial vision systems, and extracts image features using image processing algorithms to identify cutting defects. Researchers have used coaxial vision systems to acquire images of the cutting front edge and analyzed the fluctuation characteristics of the cutting front edge to identify overheating and slag defects—overheating defects are characterized by simultaneous upward jumps at both the upper and lower edges of the cutting front edge, while slag defects are characterized by upward jumps only at the lower edge. Other studies have indirectly reflected the changing trend of the cutting surface roughness by analyzing the pixel distribution of different brightness bands in spark shower coaxial vision images. In recent years, with the development of machine learning technology, some studies have begun to apply deep learning algorithms such as convolutional neural networks to laser cutting quality assessment, using the LBP algorithm to extract laser cutting surface features and using models such as support vector machines to classify cutting parameters and cutting quality.

[0034] However, the aforementioned existing technologies still have the following obvious defects and shortcomings.

[0035] First, manual inspection methods are inefficient and highly subjective. Manual inspection not only requires personnel with specialized knowledge, but also suffers from visual fatigue as working hours increase, leading to a significant decrease in accuracy. Furthermore, manual sampling can only cover a portion of products, failing to achieve full inspection and posing a risk of missed inspections. In addition, the speed of manual inspection is difficult to match the production pace of high-speed laser cutting lines, becoming a bottleneck restricting production efficiency.

[0036] Second, existing machine vision-based image detection methods are too simplistic and coarse in their representation of image features. Current methods largely focus on the presence or absence of fluctuations at the cutting edge or the overall change in spark burst brightness, lacking refined and quantitative extraction of various quality features such as slag adhesion, overheating, roughness, and texture trends. For example, for slag adhesion defects, existing visual algorithms can only determine "whether slag is present," but cannot quantify key parameters such as the projected area ratio, adhesion height, and spatial distribution density of the slag, let alone distinguish between different slag morphologies such as continuous and discrete slag adhesion. For overheating defects, existing methods can only identify the occurrence of overheating, but cannot quantify the width trend of the heat-affected zone or the area range of the overheated region. For roughness detection, existing methods mostly rely on indirect inference from side-view images or spark burst images, lacking quantitative analysis methods directly based on cross-sectional texture images.

[0037] Third, existing technologies lack a deep mapping relationship between image features and the physical mechanisms of laser processing. Laser cutting is essentially a multiphase dynamic equilibrium process involving the conversion of light energy into heat energy, the melting of solid metal, and the sliding of molten metal driven by gas flow. The cutting quality is the result of the combined effects of multiple physical field parameters such as energy density, gas flow intensity, and material thickness. However, most existing methods only remain at the superficial level of "extracting image features and identifying defect types," failing to establish a quantitative mapping relationship between image features (such as slag morphology, overheating degree, roughness value, texture tilt angle, etc.) and multiple physical field parameters (such as actual energy density, gas flow peeling intensity, multiphase flow conversion efficiency, etc.). Furthermore, this mapping relationship has not been extended to the quantitative adjustment of laser processing parameters (such as laser power, processing speed, auxiliary gas flow rate, etc.).

[0038] Fourth, existing technologies lack a systematic framework for constructing a quality classification feature model library and a theoretical data model benchmark library. Most existing research and patents develop detection algorithms for single defect types (such as slag adhesion or overheating only), or train simple classification models using only a small number of samples. A systematic model library that integrates image features and physical mechanisms, covering multiple dimensions of quality characteristics such as slag adhesion, overheating, roughness, and texture trends, has not yet been established. This makes it difficult to establish a quantitative correlation between detection results and specific process adjustment schemes, hindering the achievement of closed-loop control from "quality inspection" to "process optimization."

[0039] In summary, existing technologies have not yet solved the technical problem of how to organically integrate the extraction of refined features from laser-cut cross-section images, the quantitative mapping of multi-physics parameters, and the quantitative adjustment of laser processing parameters, making it difficult to meet the urgent needs of modern intelligent laser cutting manufacturing for online quality inspection and closed-loop process control.

[0040] To address the aforementioned issues, this application proposes a method for constructing a model library.

[0041] Please see Figure 1 , Figure 1 This is a flowchart illustrating a model library construction method provided in an embodiment of this application. Figure 1 As shown, it includes the following steps: Step S101: Preprocess the acquired cross-sectional image of the laser-cut workpiece and extract the region of interest.

[0042] In one embodiment, the preprocessing of the acquired laser-cut workpiece cut cross-section image to extract the region of interest includes: performing adaptive filtering on the cut cross-section image to obtain a filtered image; performing histogram equalization on the filtered image to obtain an enhanced image; and performing edge detection on the enhanced image to segment the cut cross-section region and the slag edge region to obtain the region of interest.

[0043] Specifically, the acquired images of the cut surface of the laser-cut workpiece are preprocessed to extract the region of interest. In the actual laser cutting process, the beam acting on the material cross-section produces complex morphological features. The real-time acquired images of the laser-cut cross-section contain stripe morphology and dynamic information of slag spray at the lower end. However, due to the strong splash noise and ambient light interference in the laser processing environment, the signal-to-noise ratio of the original image is low, and preprocessing is required to improve image quality.

[0044] First, the real-time acquired cut cross-section image undergoes adaptive filtering. The adaptive filtering algorithm dynamically adjusts the filtering parameters based on the noise distribution characteristics of local image regions, effectively eliminating high-frequency noise generated by laser spatter while preserving as much detail as possible in the cross-section stripes and slag edges, resulting in a filtered image. Next, histogram equalization is applied to the filtered image. By remapping the distribution of image pixel grayscale values, the dynamic range of grayscale values ​​in low-contrast areas is stretched, making the grayscale differences between the cross-section stripes, slag areas, and background material areas more significant. This further enhances the overall image contrast and eliminates uneven brightness caused by ambient light interference, resulting in an enhanced image. Finally, edge detection is performed on the enhanced image. Edge detection operators identify pixel locations where grayscale values ​​change drastically, accurately segmenting the contour boundaries of the cut cross-section area and the edge contours of the slag area, thereby extracting the region of interest containing information about the cut cross-section and slag from the entire image. Building upon this, thickness analysis can be performed on the extracted region of interest. Based on the pixel distance between the upper and lower edges of the cross-section and image calibration parameters, the material's thickness parameters are calculated. Simultaneously, the slag thickness parameters are calculated based on the relative positional relationship between the slag-coated edge and the cross-section edge. These thickness parameters provide fundamental data support for subsequent slag feature quantification and multiphysics mapping.

[0045] Step S102: Extract and quantize image features from the region of interest to obtain image feature parameters.

[0046] In one embodiment, image feature extraction and quantization are performed on the region of interest to obtain image feature parameters, including: edge contour tracking of the region of interest, extraction of the projected area ratio, adhesion height, and spatial frequency of the slag, and differentiation between continuous and discrete slag to obtain slag feature parameters; extraction of the width trend of the heat-affected zone in the region of interest and quantification of the degree of overheating to obtain overheating feature parameters; periodic frequency analysis of the region of interest using a gray-level co-occurrence matrix to obtain roughness feature parameters; measurement of the inclination angle of the cutting stripes in the region of interest relative to the normal direction of the plate, and fitting of the trend curve equations of each layer of texture using the least squares method to obtain texture trend feature parameters; the image feature parameters are composed of slag feature parameters, overheating feature parameters, roughness feature parameters, and texture trend feature parameters.

[0047] Specifically, image features are extracted and quantified from the region of interest to obtain image feature parameters. This step is one of the core innovations. For the four types of image features involved in the quality of laser cutting process—slag adhesion, overheating, roughness, and texture trends—targeted image processing algorithms are used for quantitative characterization, transforming subjective visual judgments into objective, calculable parameters.

[0048] In terms of slag feature extraction, the edge contour of the region of interest is first tracked, and the closed contour is tracked pixel by pixel along the boundary of the slag region. Based on this, the projection area ratio of the slag region on the cross section (i.e., the percentage of the slag projection area to the total cross section area), the adhesion height of the slag (i.e., the maximum extension distance of the slag along the thickness direction of the plate) and the spatial frequency (i.e., the distribution density of slag particles per unit length, used to characterize the density of slag particles) are calculated.

[0049] Furthermore, based on the morphological characteristics of the slag-laden areas, if the slag is spatially continuous with no obvious intervals between areas, it is classified as continuous slag; if the slag consists of dispersed, isolated particles with a certain distance between them, it is classified as discrete slag. These slag-laden characteristic parameters comprehensively reflect the severity and distribution pattern of the slag, providing quantitative input for the subsequent construction of a slag quality prediction model. Traditional visual algorithms only determine "whether or not slag is present," while this solution achieves a refined characterization of the slag-laden state through the aforementioned multi-dimensional quantitative features.

[0050] In terms of overheating feature extraction, the width trend of the heat-affected zone (HAZ) in the region of interest is extracted. During laser cutting, the width of the HAZ may vary along the thickness direction of the sheet material. By measuring the lateral expansion width of the HAZ layer by layer along the thickness direction, the trend curve of the HAZ width variation can be obtained. Based on this, the degree of overheating is classified and determined for corner areas (i.e., the turning points where the cutting direction changes) and continuous cutting areas, respectively. Specifically, the degree of overheating is quantified by detecting whether the width of the HAZ exceeds the normal threshold and whether there is melting collapse or protrusion on the material surface. The above extraction results are used to subsequently evaluate the distribution characteristics of the beam energy in the material and the overheating phenomenon caused by the beam characteristics, laying the foundation for the construction of the energy density mapping sub-model.

[0051] In terms of roughness feature extraction, a gray-level co-occurrence matrix (GLCM) is used to perform periodic frequency analysis on the region of interest. The GLCM effectively characterizes the microstructural features of image texture, such as roughness, periodicity, and directionality, by statistically analyzing the occurrence probability of pixel gray-level pairs at specific directions and distances in the image. By extracting second-order statistics such as contrast, correlation, energy, and homogeneity from the GLCM, the periodic frequency distribution of cross-sectional stripes is analyzed, yielding the roughness stripe identification results of the cross-sectional surface. This transforms the cross-sectional roughness into a computable feature parameter.

[0052] In terms of texture trend feature extraction, the inclination angle of the cutting stripes in the region of interest relative to the normal direction of the sheet material is accurately measured. Cutting stripes are microscopic morphologies formed by the dragging of molten material along the cutting direction during laser cutting driven by airflow. Their inclination angle reflects the interaction between process parameters such as cutting speed and airflow intensity and the material. Based on this, the least squares method is used to perform curve fitting on texture stripes of different levels, constructing trend curve equations for each level of texture. The slope and curvature of the trend curves reflect the variation law of the cutting stripes' direction along the sheet thickness.

[0053] The image feature parameters are composed of the above-mentioned slag-coating feature parameters, overheating feature parameters, roughness feature parameters, and texture trend feature parameters, which serve as the input for establishing the first quantitative correlation in the subsequent process.

[0054] Step S103: Establish the first quantization correlation between image feature parameters and multiphysics parameters to obtain the first mapping model.

[0055] In one embodiment, establishing a first quantitative correlation between image feature parameters and multiphysics parameters to obtain a first mapping model includes: performing energy density correlation analysis on overburning feature parameters and texture trend feature parameters to establish a first sub-mapping relationship between the actual energy density absorbed by the material per unit volume and the overburning feature parameters and texture trend feature parameters, thus obtaining an energy density mapping sub-model; performing airflow stripping intensity correlation analysis on texture trend feature parameters to establish a second sub-mapping relationship between airflow stripping intensity and boundary layering position and texture trend feature parameters, thus obtaining an airflow intensity mapping sub-model; evaluating multiphase flow conversion efficiency based on the energy density mapping sub-model and the airflow intensity mapping model to establish a third sub-mapping relationship between multiphase flow conversion efficiency and actual energy density and airflow stripping intensity, thus obtaining a multiphase flow conversion efficiency mapping sub-model; and constructing a first mapping model from the energy density mapping sub-model, the airflow intensity mapping model, and the multiphase flow conversion efficiency mapping model.

[0056] Specifically, a first quantitative correlation is established between image feature parameters and multiphysics parameters to obtain a first mapping model. The physical essence of the laser cutting process is a multiphase flow dynamic equilibrium process involving the conversion of light energy into heat energy, the melting of solid metal, and the sliding of molten metal driven by airflow. The cutting quality directly depends on the efficiency of this multiphase flow conversion. Therefore, this step focuses on three core multiphysics dimensions—multiphase flow conversion efficiency, energy density, and airflow intensity—to establish a quantitative mapping relationship with image feature parameters, achieving quantitative inference from "image features" to "physical field state."

[0057] First, an energy density mapping sub-model is established. Energy density correlation analysis is performed on overheating characteristic parameters and texture trend characteristic parameters to establish a first sub-mapping relationship between the actual energy density absorbed by the material per unit volume and the overheating and texture trend characteristic parameters. In actual processing, when the energy density absorbed by the material per unit volume exceeds the theoretical threshold required for melting and vaporization, excess energy will lead to overheating; while the inclination angle and texture morphology of the cutting stripes reflect the uniformity of energy distribution along the material thickness direction. Based on overheating characteristic parameters (such as heat-affected zone width and overheating area ratio) and texture trend characteristic parameters (such as stripe inclination angle and trend curve slope), combined with laser processing parameters (such as power and spot radius), the system calculates the actual energy density absorbed by the material per unit volume in real time, thereby establishing a mapping relationship between image characteristic parameters and energy density, resulting in the energy density mapping sub-model.

[0058] Secondly, an airflow intensity mapping sub-model is established. Correlation analysis of airflow peeling intensity is performed on texture trend characteristic parameters to establish a second sub-mapping relationship between airflow peeling intensity, boundary layering position, and texture trend characteristic parameters. During laser cutting, the kinetic energy carried by the auxiliary gas is the main driving force for the molten material to peel off and be discharged from the kerf. The airflow peeling intensity directly affects the discharge efficiency of the molten material and the morphology of residual slag inside the kerf. The inclination angle of the cutting stripes, the stripe spacing, and their variation trend along the thickness direction collectively reflect the peeling intensity of the airflow at different depths within the kerf, as well as the layering position of the interaction between the airflow boundary layer and the material. By performing regression fitting between the texture trend characteristic parameters and measured data of airflow peeling intensity, a mapping relationship for inferring airflow intensity from texture features can be established, resulting in the airflow intensity mapping sub-model.

[0059] Based on this, the multiphase flow conversion efficiency is evaluated using an energy density mapping sub-model and an airflow intensity mapping sub-model. Multiphase flow conversion efficiency reflects the overall energy and mass transfer efficiency of the laser light energy conversion into heat energy, heat energy transfer to the material, and the effective discharge of molten material from the cut by the airflow. The system comprehensively evaluates the actual energy density value output by the energy density mapping sub-model and the airflow peeling intensity value output by the airflow intensity mapping sub-model, establishing a third sub-mapping relationship between multiphase flow conversion efficiency and actual energy density and airflow peeling intensity. That is, when both energy density and airflow intensity are within the optimal matching range, the multiphase flow conversion efficiency is the highest, and the corresponding cutting quality is optimal; conversely, if the energy density is too high and the airflow peeling intensity is insufficient, it will manifest as overburning accompanied by thickened slag; if the energy density is moderate but the airflow intensity is too high, it may lead to overcutting or stripe disorder.

[0060] The first mapping model is constructed by the energy density mapping sub-model, the airflow intensity mapping sub-model, and the multiphase flow conversion efficiency mapping sub-model. This first mapping model realizes the cross-domain mapping from "image feature parameters" to "multi-physics parameters", providing a bridge for subsequent reverse deduction from multi-physics parameters to processing parameters.

[0061] Step S104: Establish the second quantitative correlation between multiphysics parameters and laser processing parameters to obtain the second mapping model.

[0062] In one embodiment, a second quantitative correlation between multiphysics parameters and laser processing parameters is established to obtain a second mapping model, including: fitting and correlating the slag-forming characteristic quantified values ​​with laser power, processing speed, and gas flow rate to construct a slag-forming correlation sub-model; performing energy density correlation analysis on the overburning characteristic parameters with laser power and spot radius to construct an overburning correlation sub-model; defining interval judgment criteria for roughness characteristic parameters and correlating them with measured texture from line-scan laser to construct a roughness correlation sub-model; performing least-squares curve fitting on texture trend characteristic parameters and extracting tangents to determine the correlation between processing speed and gas flow rate to construct a texture correlation sub-model; and constructing a second mapping model from the slag-forming correlation sub-model, the overburning correlation model, the roughness correlation model, and the texture correlation model.

[0063] Specifically, a second quantitative correlation is established between multiphysics parameters and laser processing parameters to obtain a second mapping model. Based on the first mapping model, this step further establishes a quantitative correlation between multiphysics parameters and adjustable laser processing parameters, forming a complete mapping link of "image features, multiphysics, and process parameters," providing a quantitative basis for process adjustment.

[0064] In constructing the slag-coating correlation sub-model, the quantified values ​​of slag-coating characteristics were fitted and correlated with laser power, processing speed, and gas flow rate. Slag formation is closely related to the matching relationship between laser energy input, cutting speed, and gas-assisted slag removal capability. The following formula is used to construct the slag-coating quality prediction sub-model: S=k1×P / Vf k2×Qi+k3 Where S is the quantitative value of slag coating characteristics (comprehensively reflecting the ratio of slag coating projected area, adhesion height and spatial frequency), P is the laser power, Vf is the processing speed, Qi is the gas flow rate, and k1, k2 and k3 are fitting coefficients.

[0065] In the above formula, P / Vf represents the laser energy input per unit cutting length (i.e., linear energy density). A higher value indicates more heat received per unit length of material. Qi represents the driving force of the auxiliary gas on the molten material during stripping. To adapt to different processing scenarios, the above fitting relationships are classified and matched by differentiating material types (e.g., carbon steel, stainless steel, aluminum alloys), plate thicknesses (e.g., thin plates, medium-thick plates, thick plates), gas types (e.g., oxygen, nitrogen, air), and beam types (e.g., continuous beams, pulsed beams). This outputs sub-models of energy density variation and airflow variation associated with slag characteristics. For example, the slag formation pattern in the oxygen cutting scenario for thick carbon steel differs significantly from that in the nitrogen cutting scenario for thin stainless steel, requiring different fitting coefficients for matching.

[0066] In constructing the overburning correlation sub-model, energy density correlation analysis was performed on the overburning characteristic parameters and laser power and spot radius. Overburning essentially occurs when the laser energy density received per unit area or unit volume of material exceeds the material's tolerance limit. Specifically, the overburning energy density range is determined by defining laser energy density boundary conditions, with different upper limit thresholds set for different materials and plate thicknesses. The actual energy density acting on the material surface is defined by output power and spot radius; energy density is directly proportional to power and inversely proportional to spot area. Simultaneously, the area of ​​surface collapse and protrusion in overburning is quantified through image features, establishing a mapping relationship between energy density and the area of ​​the overburned region. When the energy density is within the normal range, the overburned region area is zero or minimal; when the energy density exceeds the overburning threshold, the overburned region area monotonically increases with increasing energy density. The resulting overburning correlation sub-model can infer whether the current actual energy density exceeds the limit based on the overburning characteristic parameters and provide suggested directions for power or spot radius adjustments.

[0067] In constructing the roughness correlation sub-model, the first step is to define the interval judgment criteria for roughness feature parameters. In laser processing, the arithmetic mean roughness Ra (the arithmetic mean of the absolute values ​​of the profile offsets within the sampling length) and peak-valley height deviation PV (the height difference between the highest and lowest peaks within the sampling length) are commonly used to define roughness judgment criteria for different spatial intervals. Specifically, the material surface is divided into three detection intervals along the thickness direction: upper, middle, and lower, corresponding to the roughness characteristics of the inlet, middle, and outlet sections of the cut, respectively. The combined values ​​of Ra and PV are used to determine whether each interval meets the quality requirements. This division method can confirm the beam action state during the process (reflected in the upper section roughness characteristics) and the overall slag removal performance (reflected in the middle and lower section roughness characteristics). Based on this, the measured texture and surface roughness detection results of the linear laser scanning are correlated. The roughness feature parameters obtained from image processing are calibrated and aligned with the measured values ​​of the roughness meter. Roughness features under different process conditions are classified and labeled (e.g., smooth, slightly rough, severely rough, etc.), constructing a roughness classification label sub-model.

[0068] In constructing the texture association sub-model, least-squares curve fitting is performed on the texture trend feature parameters. During the cutting process, the molten material, dragged by the airflow, forms drag lines on the cutting surface, namely upper and lower texture lines. Their direction and shape are jointly controlled by the processing speed and airflow state. Specifically, least-squares fitting is performed on the material processing layers (i.e., dividing the plate thickness into multiple layers) and the drag lines formed by the upper and lower texture lines. The fitted texture curve reflects the variation law of texture in the depth direction. Based on this, the tangent of the fitted curve is extracted, that is, the slope of the tangent at each point on the curve is calculated. The direction of the tangent represents the instantaneous direction of the molten material being dragged out by the airflow at that depth position. According to the deflection angle and rate of change of the tangent, the correlation between processing speed and gas flow rate and texture features is determined: when the processing speed is faster, the texture tilt angle increases and the stripes become finer; when the gas flow rate increases, the stripe tilt may intensify or become disordered. The texture association sub-model established in this way can infer the matching state of the current processing speed and airflow through texture image features.

[0069] A second mapping model is constructed by combining the slag-coating correlation sub-model, the overheating correlation sub-model, the roughness correlation sub-model, and the texture correlation sub-model. This second mapping model realizes the quantitative mapping from "multi-physics parameters" to "laser processing parameters". Combined with the first mapping model, a complete technical link is formed from "image feature parameters" to "multi-physics parameters" and then to "laser processing parameters".

[0070] Step S105: Construct a quality classification feature model library based on the first mapping model and the second mapping model.

[0071] In one embodiment, a quality classification feature model library is constructed based on a first mapping model and a second mapping model, including: performing offline machine learning training on the first mapping model and the second mapping model to obtain a theoretical data model benchmark library; performing association matching between image feature parameters and the theoretical data model benchmark library to obtain association results; and performing fusion optimization on the first mapping model and the second mapping model based on the association results to obtain the quality classification feature model library.

[0072] Specifically, a quality classification feature model library is constructed based on the first and second mapping models. This step involves training and fusing the aforementioned two mapping models using machine learning to form a model library system capable of classifying and determining the quality of laser-cut workpieces.

[0073] The first and second mapping models are trained offline using machine learning algorithms. In this embodiment, a deep neural network is used as the training architecture. A large number of laser cutting samples under different process conditions (covering different materials, plate thicknesses, gas types, beam types, and different combinations of process parameters) are collected. The processing flow described in steps S101 to S104 is executed sequentially on each sample to obtain the image feature parameters, the output of the first mapping model, and the output of the second mapping model for each sample, forming a labeled dataset. The deep neural network takes the image feature parameters as input and the corresponding multiphysics parameters and process parameters as supervision labels. It continuously optimizes the network weights through the backpropagation algorithm, learning the nonlinear mapping relationship from image features to physical field states and from physical field states to process parameters. After training, the knowledge learned by the deep neural network is solidified and saved in the form of a parameterized model, forming a theoretical data model benchmark library. The essence of this benchmark library is a set of mathematical expressions representing the ternary mapping relationship of "image features - physical field parameters - process parameters", which includes the standard correspondence between various quality features and process parameters under different processing scenarios.

[0074] Subsequently, the image feature parameters are correlated and matched with the theoretical data model benchmark library. The image feature parameters of the sample to be analyzed are input into the theoretical data model benchmark library, and the degree of matching between the current image feature parameters and each standard model in the benchmark library is calculated by similarity measurement algorithms (such as Euclidean distance, cosine similarity, or Mahalanobis distance). The most similar standard model and its corresponding combination of physical field parameters and process parameters are found to obtain the feature-benchmark correlation result.

[0075] Based on the correlation results, the first mapping model and the second mapping model are fused and optimized. The actual image feature parameters of the current sample, the multiphysics parameters output by the first mapping model, the process parameters output by the second mapping model, and the standard parameters matched from the theoretical data model benchmark library are comprehensively compared to calculate the deviation vector. Based on the deviation, the model parameters are incrementally learned or fine-tuned to enable the model to continuously adapt to new processing conditions, and finally, a quality classification feature model library is obtained.

[0076] The quality classification feature model library constructed using the above method establishes a fundamental relationship between the image feature model library and the theoretical data model benchmark library. It enables image analysis and classification of image features related to laser cutting process quality, such as slag adhesion, overheating, roughness, and texture trends. This is coupled with multi-physics parameters such as actual energy density, airflow intensity, and thickness, as well as equipment fundamental parameters like multiphase flow conversion efficiency and processing technology adjustments. In practical applications, simply acquiring the cut cross-sectional image of the workpiece and extracting its feature parameters, then inputting it into the quality classification feature model library, allows for rapid matching and matching of the corresponding quality classification results and recommended theoretical process parameters. This enables rapid determination of process quality and rapid evaluation of the processing technology. This quality classification feature model library provides the foundation for subsequent closed-loop quality control adjustments of the entire cutting system and can be further integrated with process recommendations and peripheral equipment fault diagnosis functions, achieving the technical effect of replacing traditional manual judgment and experience-based analysis processes and optimizing automated process inspection.

[0077] In one embodiment, after fusing and optimizing the first mapping model and the second mapping model based on the association results to obtain a quality classification feature model library, the method further includes: matching the acquired image feature parameters of the laser-cut workpiece to be tested with the quality classification feature model library to obtain a quality classification result; and adjusting the laser processing process parameters in the second mapping model according to the quality classification result to obtain a process path evaluation recommendation scheme.

[0078] Specifically, the first step is to acquire an image of the cross-sectional area of ​​the laser-cut workpiece to be tested. In a real production environment, cross-sectional images of the workpiece after the cutting process has been completed can be acquired in real time using an online image acquisition device, or cross-sectional image samples of the cut workpiece can be acquired through offline detection methods.

[0079] Then, the same processing flow as steps S101 to S102 is performed on the cut section image to be tested. Specifically, this includes: performing adaptive filtering and histogram equalization on the cut section image to be tested to eliminate noise and ambient light interference during image acquisition, resulting in an enhanced image to be tested; performing edge detection on the enhanced image to be tested to segment the cut section area and the slag edge area, and extracting the region of interest; performing edge contour tracking on the region of interest to be tested to extract the projected area ratio, adhesion height, and spatial frequency of the slag, and distinguishing between continuous and discrete slag, thus obtaining the slag feature parameters; extracting the width trend of the heat-affected zone in the region of interest and quantifying the degree of overheating, thus obtaining the overheating feature parameters; using a gray-level co-occurrence matrix to perform periodic frequency analysis on the region of interest to be tested, thus obtaining the roughness feature parameters; measuring the inclination angle of the cutting stripes in the region of interest to be tested relative to the normal direction of the plate and using the least squares method to fit and construct a trend curve equation, thus obtaining the texture trend feature parameters; the above four feature parameters together constitute the feature parameters of the image to be tested.

[0080] Next, the feature parameters of the image to be tested are input into the quality classification feature model library for matching. The quality classification feature model library stores multiple standard quality classification models trained offline and their corresponding theoretical data model benchmarks. Each standard model corresponds to a specific set of image feature parameter ranges, physical field parameter ranges, process parameter ranges, and corresponding quality grade labels. The system calculates the matching score between the feature parameters of the image to be tested and each standard model in the quality classification feature model library using a similarity measurement algorithm (e.g., but not limited to Euclidean distance, Mahalanobis distance, or cosine similarity), and selects the standard model with the highest matching score as the matching result. Based on the matching result, the quality classification result of the workpiece to be tested is obtained. This quality classification result includes slag quality grade (e.g., no slag, slight slag, severe slag), overheating quality grade (e.g., no overheating, slight overheating, severe overheating), roughness quality grade (e.g., smooth, slightly rough, severely rough), and texture trend quality grade (e.g., normal texture, disordered texture), and can comprehensively give an overall quality evaluation grade (e.g., excellent, good, medium, poor) based on the grades of the above four dimensions. By associating and matching the feature parameters of the image under test with the theoretical data model benchmark library, the specific quality status of the current workpiece under test in terms of slag adhesion, overheating, roughness, and texture trend can be obtained quickly.

[0081] Based on this, the laser processing parameters in the second mapping model are matched and adjusted according to the quality classification results to obtain a recommended process path evaluation scheme. Specifically, when the quality classification results show that one or more dimensions have not reached the preset quality threshold, the system feeds back the deviation information in the quality classification results (i.e., the deviation vector between the measured feature parameters and the benchmark model) to the second mapping model. The second mapping model, based on the direction and magnitude of the deviation vector, reverse-engineers the direction and amount of adjustment to the laser processing parameters. For example, when the slag quality level is "severe slag," if the image feature parameters indicate that the slag projection area ratio exceeds the benchmark value and the adhesion height is significantly increased, then according to the fitting relationship S=k1×P / Vf of the slag association sub-model... Using k2×Qi+k3 to inversely calculate, when the S value is too high, it is recommended to reduce the laser power P, increase the processing speed Vf to reduce the linear energy input per unit length, or appropriately increase the gas flow rate Qi to enhance the airflow's driving force on the molten material's peeling, thereby adjusting the S value to the normal range. When the overburning quality level is "severe overburning," based on the energy density mapping relationship of the overburning correlation sub-model, it is recommended to reduce the laser power or increase the spot radius to reduce the actual energy density acting on the material surface, or adjust the focal position to make the energy distribution more uniform, so that the energy density returns to the preset overburning energy density. Within the boundary conditions; when the roughness quality level is "severe roughness", based on the judgment criteria of Ra and PV in the roughness correlation sub-model, combined with the specific roughness values ​​of different intervals (upper / middle / lower), it is recommended to adjust the matching relationship between laser power and processing speed to optimize the beam action state and slag removal performance, so that the roughness values ​​of each interval fall back to within the preset judgment criteria; when the texture trend quality level is "texture disorder", based on the deflection angle and rate of change of the tangent of the least square fitting curve in the texture correlation sub-model, it is recommended to adjust the processing speed or gas flow rate to make the texture tilt angle return to the normal range.

[0082] Finally, based on the adjustment suggestions from all the above dimensions, a process path evaluation recommendation scheme is generated. This recommendation scheme includes at least: the recommended adjustment direction and magnitude for each laser processing process parameter, and an evaluation of the expected quality improvement effect after the adjustment. This process path evaluation recommendation scheme can be output to the operator or automatically fed back to the control system of the laser cutting equipment. The control system then automatically executes closed-loop adjustment operations of the process parameters according to the recommendation scheme, thereby replacing the traditional manual judgment and experience analysis process, realizing the optimization of automated process inspection links and closed-loop adjustment of the cutting system's quality control. Furthermore, the system can continuously collect new sample data during subsequent production processes, incrementally learning and iteratively optimizing the quality classification feature model library, continuously improving the model's classification accuracy and recommendation accuracy.

[0083] This application provides a method for constructing a model library. The method includes: preprocessing an acquired image of the cut surface of a laser-cut workpiece to extract a region of interest; extracting and quantizing image features from the region of interest to obtain image feature parameters; establishing a first quantized correlation between the image feature parameters and multi-physics parameters to obtain a first mapping model; establishing a second quantized correlation between the multi-physics parameters and laser processing parameters to obtain a second mapping model; and constructing a quality classification feature model library based on the first and second mapping models. This application establishes a first quantized correlation between image feature parameters and multi-physics parameters, as well as a second quantized correlation between multi-physics parameters and process parameters such as laser power, processing speed, and auxiliary gas flow rate. This forms a complete technical link from image features to physical field state to process parameter adjustment, enabling quality inspection results to directly drive quantitative adjustment of process parameters, thereby improving the efficiency and effectiveness of quality inspection.

[0084] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0085] The following are device embodiments of this application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0086] Figure 2 This diagram illustrates a model library construction apparatus according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. The model library construction apparatus includes: The preprocessing module 201 is used to preprocess the acquired cross-sectional image of the laser-cut workpiece and extract the region of interest. Feature extraction module 202 is used to extract and quantize image features from the region of interest to obtain image feature parameters; The first mapping module 203 is used to establish a first quantization correlation between image feature parameters and multi-physics parameters to obtain a first mapping model; The second mapping module 204 is used to establish a second quantitative correlation between multiphysics parameters and laser processing parameters to obtain a second mapping model. Model library construction module 205 is used to construct a quality classification feature model library based on the first mapping model and the second mapping model.

[0087] In one embodiment, the preprocessing module 201 is further configured to perform adaptive filtering on the cut cross-section image to obtain a filtered image; The filtered image is then subjected to histogram equalization to obtain the enhanced image. Edge detection is performed on the enhanced image to segment the cut surface region and the slag edge region, thus obtaining the region of interest.

[0088] In one embodiment, the feature extraction module 202 is further used to perform edge contour tracking on the region of interest, extract the projected area ratio, adhesion height and spatial frequency of the slag, and distinguish between continuous slag and discrete slag to obtain slag feature parameters. Extract the width trend of the heat-affected zone in the region of interest and quantify the degree of overheating to obtain overheating characteristic parameters; Periodic frequency analysis of the region of interest is performed using the gray-level co-occurrence matrix to obtain roughness characteristic parameters; The angle of inclination of the cutting stripes in the region of interest relative to the normal direction of the board is measured, and the trend curve equations of each layer of texture are constructed by fitting the least squares method to obtain the texture trend characteristic parameters. Image feature parameters are composed of slag-coating feature parameters, overheating feature parameters, roughness feature parameters, and texture trend feature parameters.

[0089] In one embodiment, the first mapping module 203 is further used to perform energy density correlation analysis on the overburning characteristic parameters and texture trend characteristic parameters, establish a first sub-mapping relationship between the actual energy density absorbed by the material per unit volume and the overburning characteristic parameters and texture trend characteristic parameters, and obtain an energy density mapping sub-model; A correlation analysis of airflow stripping intensity is performed on the texture trend feature parameters to establish a second sub-mapping relationship between airflow stripping intensity, boundary layering position and texture trend feature parameters, thus obtaining the airflow intensity mapping sub-model. Multiphase flow conversion efficiency is evaluated based on energy density mapping sub-model and airflow intensity mapping sub-model. A third sub-mapping relationship is established between multiphase flow conversion efficiency and actual energy density and airflow stripping intensity, resulting in a multiphase flow conversion efficiency mapping sub-model. The first mapping model is constructed by the energy density mapping sub-model, the airflow intensity mapping sub-model, and the multiphase flow conversion efficiency mapping sub-model.

[0090] In one embodiment, the second mapping module 204 is further used to fit and correlate the quantified value of slag adhesion characteristics with laser power, processing speed and gas flow rate to construct a slag adhesion correlation sub-model; Energy density correlation analysis was performed on the overburning characteristic parameters with laser power and spot radius to construct an overburning correlation sub-model; A roughness feature parameter interval judgment standard is defined and associated with the measured texture by line-scan laser to construct a roughness association sub-model. Least square curve fitting is performed on the texture trend feature parameters and the tangent is extracted to determine the correlation between processing speed and gas flow rate, and a texture correlation sub-model is constructed. A second mapping model is constructed from the slag-attached correlation sub-model, the overheating correlation sub-model, the roughness correlation sub-model, and the texture correlation sub-model.

[0091] In one embodiment, the model library construction module 205 is further used to perform offline machine learning training on the first mapping model and the second mapping model to obtain a theoretical data model benchmark library; The image feature parameters are correlated and matched with the theoretical data model benchmark library to obtain the correlation results; Based on the correlation results, the first mapping model and the second mapping model are fused and optimized to obtain a quality classification feature model library.

[0092] In one embodiment, based on the association results, after the model library construction module 205, it further includes an evaluation module, which is used to match the acquired image feature parameters of the laser-cut workpiece under test with the quality classification feature model library to obtain the quality classification result; Based on the quality classification results, the laser processing parameters in the second mapping model are matched and adjusted to obtain a recommended scheme for process path evaluation.

[0093] This application provides a model library construction device, specifically used for: preprocessing acquired images of the cut surface of a laser-cut workpiece to extract regions of interest; extracting and quantizing image features from the regions of interest to obtain image feature parameters; establishing a first quantized correlation between the image feature parameters and multi-physics parameters to obtain a first mapping model; establishing a second quantized correlation between the multi-physics parameters and laser processing parameters to obtain a second mapping model; and constructing a quality classification feature model library based on the first and second mapping models. This application establishes a first quantized correlation between image feature parameters and multi-physics parameters, and a second quantized correlation between multi-physics parameters and process parameters such as laser power, processing speed, and auxiliary gas flow rate, forming a complete technical chain from image features to physical field states and then to process parameter adjustment. This enables quality inspection results to directly drive quantitative adjustments of process parameters, improving the efficiency and effectiveness of quality inspection.

[0094] This application Figure 3 A schematic diagram of a computer device is provided. (Example) Figure 3 As shown, the computer device 3 in this embodiment includes a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various model library construction method embodiments described above, for example... Figure 1 Steps 101 to 105 are shown. Alternatively, when processor 301 executes computer program 303, it implements the functions of each module / unit in the above-described embodiments of the model library construction apparatus, for example... Figure 2 The functions of modules / units 201 to 205 are shown.

[0095] This application also provides a readable storage medium storing a computer program, which, when executed by a processor, is used to implement the model library construction method provided in the various embodiments described above.

[0096] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0097] This application also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the execution of the executable instructions by the at least one processor causes the device to implement the model library construction methods provided in the various embodiments described above.

[0098] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0099] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a model library, characterized in that, include: The acquired images of the cut surface of the laser-cut workpiece are preprocessed to extract the region of interest; Image feature extraction and quantization are performed on the region of interest to obtain image feature parameters; Establish a first quantized correlation between the image feature parameters and the multiphysics parameters to obtain a first mapping model; A second quantitative correlation between the multiphysics parameters and laser processing parameters is established to obtain a second mapping model; A quality classification feature model library is constructed based on the first mapping model and the second mapping model.

2. The model library construction method as described in claim 1, characterized in that, The preprocessing of the acquired cross-sectional image of the laser-cut workpiece to extract the region of interest includes: The cut cross-sectional image is subjected to adaptive filtering to obtain the filtered image; The filtered image is then subjected to histogram equalization to obtain the enhanced image; Edge detection is performed on the enhanced image to segment the cut surface region and the slag edge region, thus obtaining the region of interest.

3. The model library construction method as described in claim 1, characterized in that, The step of extracting and quantizing image features from the region of interest to obtain image feature parameters includes: Edge contour tracking is performed on the region of interest to extract the projected area ratio, adhesion height and spatial frequency of the slag, and to distinguish between continuous and discrete slag to obtain slag characteristic parameters. Extract the width trend of the heat-affected zone in the region of interest and quantify the degree of overheating to obtain overheating characteristic parameters; Periodic frequency analysis of the region of interest is performed using the gray-level co-occurrence matrix to obtain roughness characteristic parameters; The angle of inclination of the cutting stripes in the region of interest relative to the normal direction of the board is measured, and the trend curve equation of each layer of texture is constructed by fitting the least squares method to obtain the texture trend feature parameters. The image feature parameters are composed of the slag-laden feature parameters, the overheating feature parameters, the roughness feature parameters, and the texture trend feature parameters.

4. The model library construction method as described in claim 3, characterized in that, The step of establishing a first quantization correlation between the image feature parameters and multiphysics parameters to obtain a first mapping model includes: Energy density correlation analysis is performed on the overburning characteristic parameters and the texture trend characteristic parameters to establish a first sub-mapping relationship between the actual energy density absorbed by the material per unit volume and the overburning characteristic parameters and the texture trend characteristic parameters, thereby obtaining an energy density mapping sub-model; A correlation analysis of airflow stripping intensity is performed on the texture trend feature parameters to establish a second sub-mapping relationship between airflow stripping intensity and boundary layering position and the texture trend feature parameters, thus obtaining an airflow intensity mapping sub-model. Based on the energy density mapping sub-model and the airflow intensity mapping sub-model, the multiphase flow conversion efficiency is evaluated, and a third sub-mapping relationship is established between the multiphase flow conversion efficiency and the actual energy density and the airflow stripping intensity, thus obtaining the multiphase flow conversion efficiency mapping sub-model. The first mapping model is constructed from the energy density mapping sub-model, the airflow intensity mapping sub-model, and the multiphase flow conversion efficiency mapping sub-model.

5. The model library construction method as described in claim 3, characterized in that, The establishment of a second quantitative correlation between the multiphysics parameters and laser processing parameters, resulting in a second mapping model, includes: The quantified values ​​of slag adhesion characteristics are fitted and correlated with laser power, processing speed and gas flow rate to construct a slag adhesion correlation sub-model; Energy density correlation analysis was performed on the overburning characteristic parameters with laser power and spot radius to construct an overburning correlation sub-model; A roughness feature parameter interval judgment standard is defined and associated with the measured texture by line-scan laser to construct a roughness association sub-model. Least square curve fitting is performed on the texture trend feature parameters and the tangent is extracted to determine the correlation between processing speed and gas flow rate, and a texture correlation sub-model is constructed. The second mapping model is constructed from the slag-attached correlation sub-model, the overheating correlation sub-model, the roughness correlation sub-model, and the texture correlation sub-model.

6. The model library construction method as described in claim 1, characterized in that, The construction of a quality classification feature model library based on the first mapping model and the second mapping model includes: Offline machine learning training is performed on the first mapping model and the second mapping model to obtain a theoretical data model benchmark library; The image feature parameters are matched with the theoretical data model benchmark library to obtain the association result; Based on the association results, the first mapping model and the second mapping model are fused and optimized to obtain the quality classification feature model library.

7. The model library construction method as described in claim 6, characterized in that, After fusing and optimizing the first mapping model and the second mapping model based on the association results to obtain the quality classification feature model library, the method further includes: The feature parameters of the acquired laser-cut workpiece image are matched with the quality classification feature model library to obtain the quality classification result; Based on the quality classification results, the laser processing parameters in the second mapping model are matched and adjusted to obtain a recommended process path evaluation scheme.

8. A model library construction apparatus, characterized in that, include: The preprocessing module is used to preprocess the acquired cross-sectional images of the laser-cut workpiece and extract the region of interest. The feature extraction module is used to extract and quantize image features from the region of interest to obtain image feature parameters; The first mapping module is used to establish a first quantized correlation between the image feature parameters and the multiphysics parameters to obtain a first mapping model. The second mapping module is used to establish a second quantitative correlation between the multiphysics parameters and the laser processing parameters to obtain a second mapping model. The model library construction module is used to construct a quality classification feature model library based on the first mapping model and the second mapping model.

9. A computer device, characterized in that, Includes a memory, and one or more processors communicatively connected to the memory; The memory stores instructions that can be executed by the one or more processors to enable the one or more processors to implement the model library construction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Includes a program or instructions that, when run on a computer, implement the model library construction method of any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the model library construction method according to any one of claims 1 to 7.

Images