Steel structural part intelligent blanking method and system based on laser cutting

Abstract

The invention relates to the technical field of laser cutting, in particular to an intelligent steel structural part blanking method based on laser cutting, which comprises the following steps: acquiring three-dimensional shape data, reflectivity data, alloy element concentration and alloy element concentration distribution of the surface of a movable arm mainboard in a cutting pre-scanning area; the reflectivity is decomposed through the alloy element concentration, and the inherent reflectivity and the deformation-induced reflectivity are obtained; and synchronizing the acquired data according to a timestamp, and generating a basic state vector of the detection point. According to the method, a multi-dimensional risk assessment system is constructed by defining a strong coupling region (first-level risk), a high warping region / element segregation region (second-level risk) and a high reflection region (third-level risk) and combining parameters such as a coupling boundary response function and an element concentration gradient; a hierarchical control strategy and feedforward prediction are designed for different region types; the key process problem in high-strength steel laser cutting is solved in a unified mode, and the precision, efficiency and stability of steel structural part blanking are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of laser cutting, and particularly to an intelligent blanking method and system for steel structure parts based on laser cutting. Background Art

[0002] Laser cutting is a technology that uses a laser beam with a high energy density to heat, melt or vaporize materials, and it can achieve high-precision and high-speed cutting of steel. Intelligent blanking of steel structure parts refers to the process of cutting raw materials (such as metal sheets, profiles, etc.) into the required shapes and sizes according to the requirements of product design drawings. In the laser cutting of steel structure parts, the physical properties of the sheet have a decisive influence on the cutting quality. During the cutting process of the boom main board of an excavator, the boom main board will have local warping due to transportation, storage or residual stress, and needs to be leveled before cutting; however, if the leveling is not thorough, the focus will still shift due to the warping of the boom main board during laser cutting, and then the distance between the laser focus and the sheet surface will fluctuate during the movement of the laser focus, which will further lead to uneven heat input during the cutting process, resulting in problems such as slag hanging, burrs, and an enlarged heat affected zone on the cutting edge, affecting the cutting quality.

[0003] The boom main board of an excavator is a core component of the load-bearing structure of the excavator and is usually made of high-strength steel; its alloy element content is relatively high, mainly including Cr, Mo, and Ni. These elements form a dense oxide film on the surface of the boom sheet, significantly increasing the reflectivity of the material to the laser; when the laser focus irradiates the surface of the boom main board, most of the energy is reflected, and the energy actually used to melt the sheet accounts for a low proportion in the input power, resulting in insufficient energy density, which will further lead to insufficient cutting depth and increased cutting time. During the cutting process, the molten metal will splash and recondense on the sheet surface, forming a local high-reflection area, resulting in fluctuations in the absorption rate; especially when the leveling of the boom main board is not thorough, the distance between the laser focus and the surface fluctuates due to the warping of the sheet, and the energy density changes accordingly. In the surface depression area of the boom main board, due to the concentration of laser energy, the amount of molten metal splashing increases, forming a dense high-reflection area; while in the convex area, due to the dispersion of energy, the splashing decreases but the oxide layer thickens; further resulting in extremely uneven reflectivity distribution on the boom main board, affecting the laser cutting effect.

[0004] Therefore, an intelligent blanking method and system for steel structure parts based on laser cutting are proposed to solve the above-mentioned problems. Summary of the Invention

[0005] Technical Problems to be Solved In view of the above-mentioned disadvantages of the prior art, the present invention provides an intelligent blanking method and system for steel structure parts based on laser cutting, which can effectively solve the problem that the fluctuation of the laser focus caused by the incomplete leveling of the boom main board in the prior art affects the cutting effect.

[0006] Technical solution To achieve the above objectives, the present invention is realized through the following technical solutions: The present invention provides an intelligent blanking method for steel structure parts based on laser cutting. The technical solutions adopted by the present invention are as follows: within the pre-scanning area of cutting, collect the three-dimensional topography data, reflectivity data, alloy element concentration and alloy element concentration distribution on the surface of the boom main board; decompose the reflectivity using the alloy element concentration to obtain the inherent reflectivity and deformation-induced reflectivity; synchronize the acquired data according to the time stamp to generate the basic state vector of the detection point; Calculate the standard deviation of reflectivity fluctuation, flatness spatial gradient and deformation reflectivity gradient at the position of the detection point based on the basic state vector; and fuse and construct them into a cutting state vector; merge the cutting state vectors of all detection points to construct a real-time cutting state matrix Based on the real-time cutting state matrix, divide the surface of the boom main board into regional types, and then determine the risk area level and its processing strategy based on the regional type; Execute the processing strategy for the laser focus according to the regional type where it is located, and in the regional type interaction area, determine the processing strategy to be preferentially executed based on the risk area level.

[0007] Further, the acquisition method of the three-dimensional topography data is as follows: Generate the point cloud data of the surface of the boom main board, fit the point cloud data into a B-spline surface; and calculate the absolute height deviation of each detection point; Construct a grid model of the surface of the boom main board, define the neighborhood based on the preset neighborhood window size, obtain the neighborhood relationship of each grid cell, and calculate the local roughness of the grid cell; generate a two-dimensional grid matrix based on the local roughness value, map the two-dimensional grid matrix into a two-dimensional grayscale image, and calculate the gradient amplitude of the local roughness, and generate a two-dimensional gradient matrix based on the gradient amplitudes of all grid cells.

[0008] Further, the acquisition method of the reflectivity data is as follows: Calculate the spectral reflectivity of each detection point on the surface of the boom main board based on the acquired time stamp t; perform moving average filtering on the spectral reflectivity of the detection point according to the preset window size to obtain the processed reflectivity of the detection point.

[0009] Further, the method of decomposing the reflectivity using the alloy element concentration is as follows: Compare the alloy element concentration on the surface of the boom main board with the preset alloy element concentration-optical parameter mapping table to look up the complex refractive index of the surface of the boom main board; Calculate the ideal surface reflectivity of the surface of the boom main board based on the complex refractive index, Based on the local roughness converted to the root mean square deviation , and calculate the roughness scattering attenuation factor using the root mean square deviation, and then calculate the intrinsic reflectivity of the boom main board in combination with the ideal surface reflectivity ; where is the ideal surface reflectivity; is the roughness scattering attenuation factor; is the exponential decay function, is the laser incident angle.

[0010] Calculate the roughness increment; convert the roughness increment into the root mean square increment of roughness, and calculate the extinction coefficient increment using the roughness increment; construct a correction formula based on the root mean square increment of roughness and the extinction coefficient increment: ; where is the deformation-induced reflectivity; is the ideal roughness of the boom main board; is the roughness increment; is the root mean square increment of roughness; k is the extinction coefficient; is the extinction coefficient increment.

[0011] Further, the standard deviation of reflectivity fluctuation is obtained as follows: Intercept the S processed reflectivities within the preset time window for the detection points according to the time stamp t, and calculate the reflectivity fluctuation sequence . is the processed reflectivity at the i-th time stamp; is the trend term obtained after processing; calculate the standard deviation of reflectivity fluctuation based on the fluctuation sequence; is the average value of the reflectivity fluctuation sequence; The flatness spatial gradient is obtained as follows: Fill the height deviation values of the grid cells in the grid model to obtain the grid height matrix; calculate the gradients of flatness in the x and y directions, as well as the gradient amplitude and gradient direction; then determine two adjacent grid cells of the current grid cell in the gradient direction along the gradient direction, and calculate the gradient amplitudes of the adjacent interpolation points, which are g1 and g2 respectively; If the gradient amplitude of the current grid cell is greater than both g1 and g2, then retain the grid cell; otherwise, suppress it; Normalize the gradient amplitudes of all retained grid cells, and output the normalized gradient amplitudes; compare the normalized gradient amplitudes with a preset empirical threshold, retain the normalized gradient amplitudes greater than the empirical threshold, and set the rest to 0; output the binary edge mask; The deformation reflectivity gradient is obtained as follows: Interpolate and fill the deformation-induced reflectivity of the grid cells in the grid model to obtain the deformation-induced reflectivity matrix; calculate the gradients of the deformation-induced reflectivity in the x and y directions, as well as the gradient magnitude; then define the coupled boundary response function in combination with the flatness spatial gradient and the deformation reflectivity gradient. The function formula is: 。

[0012] Further, the region types include strong coupling regions, high warpage regions, element segregation regions, and high reflection regions: Divide the strong coupling region into first-level risk regions, and the expression of the trigger condition is: ; where Ra is the local roughness; G is the gradient magnitude of the roughness; is the roughness threshold; is the roughness gradient threshold; is the strong coupling boundary threshold; Divide the high warpage region into second-level risk regions, and the expression of the trigger condition is: ; where is the absolute height deviation; is the height deviation threshold, is the flatness gradient threshold; Divide the element segregation region into second-level risk regions; the expression of the trigger condition is: ; where is the concentration of alloy element i; is the alloy element concentration threshold of alloy element i; is the concentration gradient vector of alloy element i; is the critical threshold of the concentration gradient; Divide the high reflection region into third-level risk regions, and the expression of the trigger condition is: ; where is the processed reflectivity; is the reflectivity fluctuation threshold.

[0013] Further, the processing strategy for the high warpage region is: adjust the laser focus position according to the absolute height deviation, and default to maintain the basic cutting power ; The processing strategy for the high reflection region is: adjust the power proportionally according to the difference between the processed reflectivity and the ideal surface reflectivity, and use the basic focus height set by the pre-scanning.

[0014] The processing strategy for the strong coupling region is: calculate the absolute height deviation in real time to adjust the laser focus, and adjust the power according to the coupling strength; predict the coupling region in advance and synchronously adjust the focus and power before the laser focus arrives The processing strategy for the element segregation region is: increase the power proportionally according to the amplitude of the element concentration exceeding the threshold.

[0015] Further, the method for predicting the coupling area in advance is as follows: Extract the boundary coordinates of the risk area from the cutting state matrix, determine the extension direction of the risk zone, and calculate the spatial distribution trend of the risk area; calculate the time advance of the laser focus reaching the risk area; and during the scanning stage, bind the focus compensation amount and power compensation amount of the risk area with the coordinates and store them in the feedforward control buffer queue; when the cutting head enters the prediction time window, read the control parameters in the buffer in advance and trigger the actuator to act.

[0016] An intelligent blanking system for steel structure parts based on laser cutting includes: A data acquisition module for collecting three-dimensional topography data, reflectivity data, alloy element concentration, and alloy element concentration distribution on the surface of the boom main board; A data processing module for processing the collected data and fusing and constructing a real-time cutting state matrix; A region division module for dividing the region type based on the real-time cutting state matrix, determining the risk area level of the region type and its processing strategy; A control execution module for executing the corresponding processing strategy based on the risk area level of the region type.

[0017] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are as follows: 1. In the present invention, multi-dimensional information such as point cloud data, roughness, and spectral reflectivity of the boom main board are synchronously collected by a laser ranging array and a spectral sensor, covering the deformation-sensitive areas around the cutting path, solving the defect that traditional single-point detection cannot capture global warping; based on B-spline surface fitting and triangular meshing models, the surface unevenness, roughness, and spatial gradient are quantified, and accurate modeling of complex deformations such as non-linear warping and molten metal splash condensation is realized, improving the adaptability of the system to uneven plates.

[0018] 2. In the present invention, the reflectivity is decomposed into intrinsic reflectivity and deformation-induced reflectivity, avoiding the problem of feature mixing in traditional hybrid parameter modeling; and through the correction of Fresnel's law, Beckmann scattering theory, and CNN model, the energy absorption characteristics of different regions are accurately calculated, providing a theoretical basis for the dynamic adjustment of laser power.

[0019] 3. In the present invention, a strong coupling region (primary risk), a high warpage region / element segregation region (secondary risk), and a high reflection region (tertiary risk) are defined. Combining parameters such as the coupling boundary response function and the element concentration gradient, a risk assessment from single-point detection to field analysis is constructed, and a multi-dimensional risk assessment system is established. In addition, for different region types, hierarchical control strategies and feed-forward predictions are designed, systematically solving the key process problems in laser cutting of high-strength steel, significantly improving the accuracy, efficiency, and stability of steel structure part blanking, and having strong engineering application value and technological innovation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic flow chart of the intelligent blanking method for steel structure parts in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] Embodiment 1 Referring to Figure 1 , a method for intelligent blanking of steel structure parts based on laser cutting is proposed in this case, including the following steps: Install a data acquisition module on the laser cutting head. The data acquisition module delimits a pre-scanning area according to the cutting path of the laser cutting head on the boom main board, covering the surrounding area of the cutting path to ensure that the warpage deformation area that may affect cutting on the surface of the boom main board can be captured. Based on the three-dimensional topography data and reflectivity data of the surface of the boom main board collected by the data acquisition module, a real-time cutting state matrix is constructed.

[0023] Based on the real-time cutting state matrix, the surface of the boom main board is divided into region types, and then the risk region level is determined based on the region type. Based on the region type level, a processing strategy is determined, and in the region type conflict area, the processing strategy to be preferentially executed is determined based on the risk region level.

[0024] The method for constructing the real-time cutting state matrix is as follows: Collect three-dimensional topography data: Scan the boom main board at a preset interval through a laser ranging array. Each sensor measures the vertical distance from the boom main board to the laser head in real time to generate point cloud data. The point cloud data is fitted into a B-spline surface, and the fitting equation is: ; where is the coordinate (x, y, z) of any point on the surface, calculated through parameters u and v; is the control point, which determines the current framework of the surface and is obtained by least squares fitting based on the point cloud data; is the p-th B-spline basis function in the u direction, and is the q-th B-spline basis function in the v direction; N and M are the numbers of control points in the u direction and v direction respectively; i and j are index variables. By quantifying the concavity and convexity of the surface of the boom main board, it can provide a geometric basis for the dynamic compensation of the laser focus during the laser cutting process, solve the problem that traditional plane fitting cannot handle non-linear warping, and improve the adaptability to the cutting of the boom main board with uneven surface.

[0025] The reference plane height of the boom main board is obtained by fitting the point cloud data using the least squares method , and the absolute height deviation of each detection point in the point cloud data is calculated based on the reference plane height ; where z is the actual measured height of the detection point in the point cloud data.

[0026] Collect reflectivity data: Based on the calibrated spectral sensor, the surface of the boom main board is detected, and the spectral reflectivity of each detection point on the surface of the boom main board is calculated based on the collected timestamp t , and the calculation formula is: ; where is the measured light intensity value at the wavelength ; is the weight coefficient at the wavelength , which is determined by fitting historical data; is the detected spectral wavelength range. The spectral reflectivities of all detection points are combined to construct a reflectivity data set, and the high-reflection area on the surface of the boom main board is identified based on the dynamic threshold method. Specifically: The spectral reflectivities of all detection points on the surface of the boom main board collected by the spectral sensor are subjected to moving average filtering according to a preset window size to eliminate high-frequency noise, and the processing formula is: ; where is the processed reflectivity of the K-th detection point; is the spectral reflectivity of the K-th detection point; W is the window size. During the process of collecting the reflectivity of the detection points, there may be sudden noise in the sensor data (such as electromagnetic interference or dust interference), resulting in inaccurate single measurement; by using the average value of several points before and after (i.e., the window size) to represent the current value, the data is made smoother and more reliable, and can better reflect the true level. Then, the processed reflectivity of the detection points is compared with the preset reflectivity reference threshold. When the processed reflectivities of 3 consecutive detection points are greater than the reflectivity reference threshold, they are marked as high-reflection areas.

[0027] It should be noted that during the acquisition of 3D topography and reflectivity data, there is a strong coupling between the surface flatness and reflectivity characteristics of the boom mainboard. This makes it difficult to accurately characterize the cutting behavior of the laser focus when constructing the real-time cutting state matrix from the directly acquired 3D topography and reflectivity data. In particular, treating the boom mainboard's flatness and reflectivity as independent parameters without separating the inherent material reflectivity (determined by the boom mainboard's material chemical composition and original surface state) from the deformation-induced reflectivity (the additional reflectivity increase caused by changes in the plate's surface geometry, including laser energy concentration causing molten metal to splash, which forms a rough oxide particle layer after condensation on the boom mainboard surface). This results in a mixed feature set in the real-time cutting state matrix, making it impossible to accurately compensate for the laser focus and adjust the power. When acquiring point cloud data for the boom mainboard surface, the surface roughness data and spatial gradient characteristics of the boom mainboard are simultaneously collected to establish a flatness-roughness correlation and precisely locate high-coupling areas where molten metal splashes and condenses due to warping.

[0028] Specifically, a mesh model of the boom mainboard surface is constructed using triangulation, and a neighborhood is defined based on a preset neighborhood window size for local analysis. The neighborhood relationship of each grid cell is obtained, and the local roughness is calculated. ; where local roughness represents the root mean square deviation between the actual measured height of the detection point in the point cloud data within the neighborhood and the reference plane height; D is the number of detection points in the local area. A two-dimensional grid matrix is generated based on the output local roughness value of each grid cell ; i and j are index variables. The two-dimensional grid matrix is mapped to a two-dimensional grayscale image. The gradient of the local roughness in the x and y directions is calculated based on the Sobel operator, and the gradient amplitude is calculated. ; Where, the gradient amplitude Represents the spatial rate of change of the roughness at the grid cell in row i and column j. A larger value indicates a more dramatic change in the roughness in the local area. A two-dimensional gradient matrix is generated based on the gradient amplitudes of all grid cells for subsequent coupling region identification and risk field construction.

[0029] The absolute deviation of the height of the detection point , the local roughness Ra and gradient amplitude G of the grid unit are compared with the preset judgment threshold, and the The grid cells of are marked as high coupling areas. is the height deviation threshold, is the roughness threshold, is the roughness gradient threshold; the judgment thresholds are calibrated through experimental data.

[0030] More specifically, for the reflectivity decoupling process, the reflectivity on the surface of the boom main board is physically decomposed into the contribution of the inherent properties of the material and the contribution induced by deformation, namely the inherent reflectivity and the deformation-induced reflectivity, so as to clarify the reasons for the change in reflectivity.

[0031] The method for obtaining the inherent reflectivity is as follows: Based on the spectrometer to collect the alloy composition of the boom main board, determine the alloy element concentration in the boom main board, including chromium concentration , molybdenum concentration and nickel concentration , and their concentration distributions. Compare the alloy element concentrations with the preset alloy element concentration - optical parameter mapping table, and look up the complex refractive index (n, k) on the surface of the boom main board, which describes the refractive and absorption characteristics of the boom main board to light; where n is the real part, which determines the propagation speed of light in the boom main board; k is the imaginary part (i.e., the extinction coefficient), which determines the attenuation rate of light.

[0032] Calculate the ideal surface reflectivity based on the Fresnel reflectivity , and the calculation formula is: . The ideal surface reflectivity is the reflectivity when ignoring the surface roughness of the boom main board, and it is the reflectivity only determined by the structure of the boom main board. The Fresnel's law has different reflectivities for p-polarized light (parallel to the incident plane) and s-polarized light (perpendicular to the incident plane), but circularly polarized light is usually used in laser cutting, so the average reflectivity is taken.

[0033] Based on the Gaussian distribution, convert the local roughness into the root mean square deviation of the surface height applicable to the optical model ; and calculate the roughness scattering attenuation factor based on the Beckmann scattering theory, which is used to correct the inherent reflectivity of the material to eliminate the influence of roughness on the reflectivity and obtain the actual reflectivity only determined by the material composition. The calculation formula is: ; in the formula, is the inherent reflectivity of the boom main board; is the roughness scattering attenuation factor, which reflects the light scattering loss caused by the microscopic undulations of the surface; is the exponential attenuation function, is the laser incident angle.

[0034] The method for the alloy element concentration - optical parameter mapping table is as follows: Prepare several groups of plates with known but different component compositions as standard specimens to cover the common alloy content range; calibrate the complex refractive index (n, k) of each specimen at different wavelengths based on the ellipsometer, use a quadratic polynomial to capture the non-linear relationship between alloy element concentrations, and use the least squares method to fit to obtain the coefficients; finally, establish the alloy element concentration - optical parameter mapping table.

[0035] Based on the concentration distribution of the detection point elements in the real-time cutting state matrix, a continuous alloy element concentration cloud map is generated by inverse distance weighted interpolation IDW. The detected element concentration data are usually discrete points and cannot directly reflect the continuous distribution characteristics of the elements on the plate surface (such as band segregation, dendritic segregation); through inverse distance weighted interpolation IDW, the discrete point data are interpolated into a continuous element concentration cloud map to reveal the spatial distribution law of the microscopic composition. It provides a high-resolution composition distribution basis for subsequent optical parameter correction, enabling the reflectivity calculation to match the local composition differences rather than relying on the overall average composition. The alloy element concentration cloud map is used as the data input of the CNN model, and the pre-trained CNN model outputs the complex refractive index correction amount, including the real part correction amount and the imaginary part correction amount . By using machine learning model training to map the element concentration cloud map to the complex refractive index correction amount, the synergistic effect between alloy elements can be automatically captured; furthermore, during the subsequent cutting process of the boom main board, it can adapt to unknown composition fluctuations. Recalculate the intrinsic reflectivity based on the real part correction amount and the imaginary part correction amount to obtain the corrected intrinsic reflectivity .

[0036] The training method of the CNN model is as follows: Construct a training dataset through the element concentration distribution data and ellipsometric spectroscopy calibration data of standard specimens; define the loss function as the mean square error, transfer the training dataset to the CNN model and output the predicted value, calculate the value of the corresponding loss function, and backpropagate the error gradient along the output layer to the input layer; according to the error gradient, use the optimization algorithm to update the network parameters; repeat the training until the CNN model converges or reaches the pre-set number of iterations, that is, the training of the CNN model is completed.

[0037] The acquisition method of the deformation-induced reflectivity is as follows: Calculate the roughness increment ; where is the ideal roughness after leveling the boom main board, representing the reference state without significant deformation. When the roughness increment is greater than 0, it means that the local roughness is higher than the reference, such as the area where molten metal splashes and condenses; when the roughness increment is 0, it means the local configuration is in the reference state; when the roughness increment is less than 0, it means the local roughness is lower than the reference. Then, based on the Gaussian distribution, the roughness increment is converted into the root mean square roughness increment . Calculate the extinction coefficient increment caused by molten metal splashing using the roughness increment ; where is an empirical coefficient. During the cutting process, when the molten metal splashes onto the surface of the boom main board and forms an oxide layer after cooling, its extinction coefficient k is significantly higher than that of the metal substrate. The more the amount of splash accumulation, the greater the thickness of the oxide layer, the higher the roughness, and the higher the increment value of the extinction coefficient. Finally, a correction formula is constructed based on the root mean square increment of roughness and the increment of extinction coefficient: ; In the formula, is the deformation-induced reflectivity.

[0038] Synchronize the three-dimensional topography data and reflectivity data according to the timestamp to generate the basic state vector of each detection point ; Calculate the standard deviation of reflectivity fluctuation , flatness spatial gradient and deformation reflectivity gradient at its position based on the basic state vector of the detection point. And combine the standard deviation of reflectivity fluctuation , flatness spatial gradient , deformation reflectivity gradient and the basic state vector to construct the cutting state vector. Combine the cutting state vectors of all detection points to construct the real-time cutting state matrix.

[0039] Specifically, the method for obtaining the standard deviation of reflectivity fluctuation is as follows: For each detection point, intercept S processed reflectivities within the preset time window according to the timestamp t, use the Savitzky-Golay filter to remove the low-frequency trend, and calculate the filtered reflectivity fluctuation sequence . is the processed reflectivity at the i-th timestamp; is the trend term obtained by the filtering algorithm for , which represents the slow component of the reflectivity change over time, that is, after removing the high-frequency noise, it reflects the inherent change of material properties (such as the cumulative effect of alloy element oxidation during the cutting process) or the long-term influence of macroscopic deformation (such as the reflectivity change caused by the overall thermal deformation of the sheet). The complex reflectivity change can be decomposed into components with clear physical meanings through the fluctuation sequence, providing key data support for the intelligent control of laser cutting. Calculate the standard deviation of reflectivity fluctuation based on the fluctuation sequence; In the formula, S is the number of sampling points within the time window; is the average value of the reflectivity fluctuation sequence.

[0040] The dynamic interference of molten metal spatter can be detected in real time based on the standard deviation of reflectivity fluctuations. During cutting, the molten metal spatter will affect the absorption and transmission of laser energy. The standard deviation of reflectivity fluctuations can reflect the fluctuation of reflectivity in real time. When the value of the standard deviation of reflectivity fluctuations is large, it indicates that there are many dynamic interference factors during the cutting process, such as frequent spattering, and the stability of the cutting process is poor at this time. By monitoring the standard deviation of reflectivity fluctuations, unstable factors in the cutting process can be detected in time, providing a basis for adjusting cutting parameters, such as adjusting the laser power or cutting speed, to reduce the generation of spatter and ensure the stable progress of the cutting process. In addition, during the cutting process, the standard deviation of reflectivity fluctuations in different regions may be different. By continuously monitoring the standard deviation of reflectivity fluctuations, according to its change trend, problems that may occur in the cutting area can be predicted in advance, and cutting parameters can be adjusted in advance. For example, when the standard deviation of reflectivity fluctuations begins to rise, appropriately reduce the laser power to avoid a decrease in cutting quality due to excessive spatter.

[0041] Spatial gradient of flatness The acquisition method is as follows: Based on the grid model on the surface of the boom main board, the height deviation values of the grid cells are filled by bilinear interpolation to obtain a regular grid height matrix. The Sobel operator is used to calculate the gradients of flatness in the x and y directions, as well as the gradient magnitude (i.e., the absolute value of the spatial gradient of flatness) and the gradient direction. Then, along the gradient direction, two adjacent grid cells of the grid cell in the gradient direction are determined, and the gradient magnitudes of the adjacent interpolation points are calculated by bilinear interpolation, which are g1 and g2 respectively; if the gradient magnitude of the current grid cell is greater than both g1 and g2 at the same time, then retain this grid cell; otherwise, suppress it. Normalize the gradient magnitudes of all retained grid cells so that their range is mapped to [0,1] for unified threshold judgment. Output the normalized gradient magnitude , where is the gradient magnitude after non-maximum suppression; is the global maximum value in the gradient magnitude matrix, that is, the gradient magnitude of the strongest edge point in the whole image; is the global minimum value in the gradient magnitude matrix, usually 0, because NMS has set non-edge points to 0. Compare the normalized gradient magnitude with a preset empirical threshold, retain the normalized gradient magnitude greater than the empirical threshold, and set the rest to 0. Finally, output a binary edge mask. Based on NMS, eliminate non-maximum points in the gradient magnitude, only retain the pixels of the local gradient maximum, and set the rest to 0, thinning the edge to the width of a single grid cell to solve the problem of edge thickening; then screen the edge candidate points after NMS, eliminate the noise points with lower gradient magnitudes, retain the true strong edges, output a binary edge mask, and clearly mark the edge positions. It provides an edge position signal with both accuracy and robustness for cutting path planning and focus dynamic compensation, which is the core technical link for realizing intelligent blanking of steel structure parts.

[0042] Precisely locate the mutation regions on the surface of the boom main board based on the spatial gradient of flatness, and accurately identify the regions where the height deviation of the sheet surface changes drastically. These regions are often high-risk areas during the cutting process. In these regions, the laser focus is prone to deviate from the ideal position, resulting in uneven input of cutting heat and cutting defects. After locating these risk regions through the spatial gradient of flatness, targeted measures can be taken. For example, when cutting to these regions, automatically adjust the position of the laser focus to ensure that the laser energy can accurately act on the sheet surface and improve the cutting accuracy. In addition, when planning the cutting path, considering the information of the spatial gradient of flatness can enable the cutting path to avoid or better handle high-risk regions. For regions with a large spatial gradient of flatness, a more refined cutting strategy can be adopted, such as reducing the cutting speed and increasing the action time of the laser energy, to ensure the cutting quality. At the same time, the cutting sequence can also be reasonably arranged according to the distribution of the spatial gradient of flatness, cutting the relatively flat regions first and then processing the complex regions to improve the overall cutting efficiency.

[0043] Deformation reflectivity gradient is obtained as follows: Based on the grid model on the surface of the boom main board, fill the deformation-induced reflectivity of the grid cells through bilinear interpolation to obtain the deformation-induced reflectivity matrix. Use the Sobel operator to calculate the gradients of the deformation-induced reflectivity in the x and y directions, as well as the gradient magnitude (i.e., the absolute value of the deformation reflectivity gradient). Then, combine the spatial gradient of flatness and the deformation reflectivity gradient to define the coupled boundary response function and quantify the spatial coupling strength between flatness and reflectivity; the function formula is: . Then, based on the coupled boundary response function, map the grid cells on the surface of the boom main board into a color image, and divide the risk regions in the image based on the strong coupling boundary threshold .

[0044] The deformation reflectivity gradient can clearly define the defect boundary and the energy absorption difference; in the cutting of the boom main board, there is a coupling relationship between flatness and reflectivity. The deformation reflectivity gradient is specifically used to locate the regions where reflectivity mutations occur due to plate deformation (such as warping), that is, the strong coupling boundary between flatness and reflectivity. The energy absorption characteristics of these regions are different from those of other regions, and problems are likely to occur during cutting. By accurately identifying these boundaries, special treatments can be carried out on these regions during cutting, such as adjusting the laser energy density, to ensure the consistency of cutting quality. And in the regions with a large deformation reflectivity gradient, due to uneven energy absorption, the input of laser energy needs to be adjusted according to its magnitude. For regions with weak energy absorption, the laser power can be appropriately increased to ensure the cutting depth; for regions with strong energy absorption, the power can be appropriately reduced to avoid over-cutting. This can make the energy input during the entire cutting process more reasonable, improve the cutting quality, and reduce the generation of cutting defects.

[0045] The method for classifying the risk region levels is as follows: Calculate the gradient amplitude of the alloy element concentration contour map; then divide the risk region based on a preset alloy element concentration threshold and a contour map gradient threshold. Among them, the alloy element concentration threshold includes the chromium concentration threshold , the molybdenum concentration threshold and the nickel concentration threshold ; when the chromium concentration is greater than the chromium concentration threshold, it will cause an increase in the density of the oxide film on the surface of the boom main board and an increase in reflectivity; when the molybdenum concentration is greater than the molybdenum concentration threshold, it means that the reflectivity of its oxide is higher than that of the metal matrix; when the nickel concentration is greater than the nickel concentration threshold, it means that a high-reflectivity oxide layer is easily formed on the surface of the boom main board and is accompanied by a change in thermal conductivity. Classify the region types: Classify the strong coupling region as a first-level risk region, indicating a strong coupling between flatness and reflectivity; the expression of the trigger condition is: ; Classify the high warping region as a second-level risk region, indicating that the surface height deviation exceeds the allowable range and the spatial gradient of flatness is significant; the expression of the trigger condition is: ; where is the flatness gradient threshold; Classify the element segregation region as a second-level risk region; it means that the concentration of any alloy element exceeds its threshold and the concentration gradient exceeds the critical value, that is, there is an abnormal reflectivity region caused by composition segregation; the expression of the trigger condition is: ; where is the concentration of alloy element i; is the alloy element concentration threshold of alloy element i; is the concentration gradient vector of alloy element i, which characterizes the change rate of concentration in space; The critical threshold of the concentration gradient is used to distinguish normal concentration changes from abnormal segregation. When there are multiple alloying element segregation conflicts, the alloying element with the higher concentration is used for judgment.

[0046] The high-reflection area is divided into three risk levels; it means that the reflectivity exceeds the benchmark value or the reflectivity fluctuates violently. The trigger condition expression is: ; is the reflectivity after processing; is the reflectivity fluctuation threshold.

[0047] The processing strategy for high warpage area is: according to the absolute height deviation detected in real time, the focus position is adjusted immediately and the adjustment value is adjusted. , The focus adjustment process is smoothed by low-pass filtering algorithm to avoid high-frequency jitter. The default is to maintain the basic cutting power. If abnormal reflectivity fluctuations are detected, the power is adjusted slightly to compensate for the difference in energy absorption. Energy concentration in laser cutting is highly dependent on focal position. Focus deviation can lead to insufficient cutting depth or over-melting. Maintaining a constant distance between the focus and the surface is essential for ensuring cutting accuracy. Warping can also be accompanied by localized reflectivity changes (e.g., surface tilt causing a change in the angle of incidence), but this is primarily due to geometric deviations. Therefore, focus compensation is the primary focus, with power adjustments only used as an auxiliary aid.

[0048] The processing strategy for high reflective areas is: according to the difference between the reflectivity after treatment and the ideal surface reflectivity, adjust the power proportionally, and adjust the value , ensuring that the energy density actually used for cutting is sufficient; among them, This is the gain coefficient between reflectivity and power, calibrated based on experimental data. Dynamically adjust laser power to compensate for insufficient energy absorption due to high reflectivity, maintaining a stable cutting depth. The pre-scan focus base height is maintained, and only high warpage areas are addressed when warpage is detected.

[0049] The strategy for handling the strong coupling zone is to calculate the absolute height deviation in real time to adjust the laser focus and adjust the power significantly according to the coupling strength; predict the coupling zone in advance and adjust the focus and power synchronously before the cutting focus is reached to reduce the error caused by the system response lag. The coupling effect will cause the dual temporal and spatial unevenness of energy density (such as local overheating and insufficient energy). The focus and power need to be adjusted synchronously to break the warping and cause spatter, which then increases the reflectivity and creates a vicious cycle of insufficient energy. The method for predicting the coupling zone in advance is: Extract the boundary coordinates of the risk area from the cutting state matrix, determine the extension direction of the risk zone through principal component analysis (PCA), and calculate the spatial distribution trend of the risk area; then, based on the time-space mapping of kinematics, calculate the time advance of the laser focus reaching the risk area; and during the scanning stage, bind the focus compensation amount and power compensation amount of the risk area to the coordinates and store them in the feedforward control buffer queue; when the cutting head enters the prediction time window, read the control parameters in the buffer in advance and trigger the action of the actuator.

[0050] The processing strategy for the element segregation area is as follows: the element concentration has a non-linear relationship with the reflectivity. Increase the power proportionally according to the amplitude by which the element concentration exceeds the threshold to compensate for the insufficient energy in the high-reflectivity area; the segregation area may be accompanied by micro-warpage caused by rolling stress, and synchronous execution of focus compensation can solve the dual defects. Element segregation is an inherent property difference of the material and cannot be solved by focus adjustment. It is necessary to dynamically adjust the power to match the local energy demand in response to the changes in reflectivity and thermal characteristics caused by the composition.

[0051] When multiple types of areas are superimposed, sort the execution of the processing strategies according to the risk area levels of the type areas, and an adaptive adjustment with greater control intensity for higher risks can be achieved by assigning weights through fuzzy logic, avoiding excessive or insufficient control caused by simple priorities.

[0052] Embodiment 2 Refer to Figure 1 , on the basis of Embodiment 1, this case proposes an intelligent blanking system for steel structure parts based on laser cutting, including The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent nesting method for steel structure parts based on laser cutting, characterized in that, Including: Within the cutting pre-scanning area, collect the three-dimensional topography data, reflectivity data, alloy element concentration, and alloy element concentration distribution on the surface of the boom main board; Decompose the reflectivity using the alloy element concentration to obtain the intrinsic reflectivity and deformation-induced reflectivity; synchronize the acquired data according to the timestamp to generate the basic state vector of the detection point; Calculate the standard deviation of reflectivity fluctuations, flatness spatial gradient, and deformation reflectivity gradient at the position of the detection point based on the basic state vector; And fuse and construct them into a cutting state vector; merge the cutting state vectors of all detection points to construct a real-time cutting state matrix Based on the real-time cutting state matrix, divide the surface of the boom main board into regional types, and then determine the risk area level and its processing strategy based on the regional type; Execute the processing strategy for the laser focus according to the regional type it belongs to, and in the regional type interaction area, determine the priority execution processing strategy based on the risk area level.

2. The intelligent blanking method for steel structure parts based on laser cutting according to claim 1, wherein: The acquisition method of the three-dimensional topography data is as follows: Generate the point cloud data of the surface of the boom main board, fit the point cloud data into a B-spline surface; and calculate the absolute height deviation of each detection point; Construct a grid model of the surface of the boom main board, define the neighborhood based on the preset neighborhood window size, obtain the neighborhood relationship of each grid cell, and calculate the local roughness of the grid cell; Generate a two-dimensional grid matrix based on the local roughness value, map the two-dimensional grid matrix to a two-dimensional grayscale image, and calculate the gradient amplitude of the local roughness. Generate a two-dimensional gradient matrix based on the gradient amplitudes of all grid cells.

3. The intelligent blanking method for steel structure parts based on laser cutting according to claim 2, characterized in that: The acquisition method of the reflectivity data is as follows: Calculate the spectral reflectivity of each detection point on the surface of the boom main board based on the acquired timestamp t; perform moving average filtering on the spectral reflectivity of the detection point according to the preset window size to obtain the processed reflectivity of the detection point.

4. The intelligent blanking method for steel structure parts based on laser cutting according to claim 3, characterized in that: The method of decomposing the reflectivity using the alloy element concentration is as follows: Compare the alloy element concentration on the surface of the boom main board with the preset alloy element concentration - optical parameter mapping table to look up the complex refractive index of the surface of the boom main board; Calculate the ideal surface reflectivity of the surface of the boom main board based on the complex refractive index, Based on the conversion of local roughness into root mean square deviation , and using the root mean square deviation to calculate the roughness scattering attenuation factor, and then combining with the ideal surface reflectivity to calculate the inherent reflectivity of the boom main board ; where is the ideal surface reflectivity; is the roughness scattering attenuation factor; is the exponential decay function, is the laser incident angle. Calculate the roughness increment; convert the roughness increment into the root mean square roughness increment, and calculate the extinction coefficient increment using the roughness increment; construct a correction formula based on the root mean square roughness increment and the extinction coefficient increment: ; where, is the deformation-induced reflectivity; is the ideal roughness of the boom main board; is the roughness increment; is the root mean square increment of roughness; k is the extinction coefficient; is the extinction coefficient increment.

5. The intelligent blanking method for steel structure parts based on laser cutting according to claim 4, wherein: The standard deviation of the reflectance fluctuation is obtained as follows: Intercept S processed reflectances within a preset time window for the detection points according to the timestamp t, and calculate the reflectance fluctuation sequence . is the processed reflectance at the i-th timestamp; is the trend term obtained after processing; Calculating the standard deviation of reflectance fluctuations based on a fluctuation sequence ; is the average value of the reflectance fluctuation sequence; The spatial gradient of flatness is obtained as follows: Fill the height deviation values of the grid cells in the grid model to obtain the grid height matrix; calculate the gradients of flatness in the x and y directions, as well as the gradient amplitude and gradient direction; then determine the two adjacent grid cells of the grid cell in the gradient direction along the gradient direction, and calculate the gradient amplitudes of the adjacent interpolation points, which are g1 and g2 respectively; If the gradient amplitude of the current grid cell is greater than both g1 and g2 at the same time, retain this grid cell; Otherwise, suppress it; Normalize the gradient amplitudes of all retained grid cells, output the normalized gradient amplitudes; compare the normalized gradient amplitudes with the preset empirical threshold, retain the normalized gradient amplitudes greater than the empirical threshold, and set the rest to 0; output the binary edge mask; Deformation reflectivity gradient is obtained by the following method: Interpolate and fill the deformation-induced reflectance of grid cells in the grid model to obtain the deformation-induced reflectance matrix; calculate the gradients of the deformation-induced reflectance in the x and y directions, as well as the gradient magnitude; then define the coupled boundary response function by combining the flatness spatial gradient and the deformation reflectance gradient, and the function formula is: .

6. The intelligent blanking method for steel structure parts based on laser cutting according to claim 5, wherein: The regional types include strong coupling area, high warping, element segregation area, and high reflection area: The strongly coupled area is divided into first-level risk areas, and the expression of the triggering condition is: ; where Ra is the local roughness; G is the gradient amplitude of the roughness; is the roughness threshold; is the roughness gradient threshold; is the strongly coupled boundary threshold; The highly warped area is divided into secondary risk areas, and the expression of the triggering condition is: ; where is the absolute height deviation; is the height deviation threshold, is the flatness gradient threshold; Divide the element segregation area into a secondary risk area; the expression of the triggering condition is: ; where is the concentration of alloying element i; is the concentration threshold of alloying element i; is the concentration gradient vector of alloying element i; is the critical threshold of the concentration gradient; The highly reflective area is divided into three levels of risk areas, and the expression of the triggering condition is: ; where is the reflectivity after processing; is the reflectivity fluctuation threshold.

7. The intelligent blanking method for steel structure parts based on laser cutting according to claim 6, characterized in that: The processing strategy for the high warpage area is as follows: Adjust the laser focus position according to the absolute height deviation, and by default, maintain the basic cutting power ; The processing strategy for the high-reflection area is as follows: According to the difference between the reflectivity after processing and the ideal surface reflectivity, the power is adjusted proportionally, and the focal base height set in the pre-scanning is continued to be used. The processing strategy for the strong coupling area is as follows: The absolute height deviation is calculated in real time to adjust the laser focus, and the power is adjusted according to the coupling strength; The coupling area is predicted in advance, and the focus and power are adjusted synchronously before the laser focus arrives. The processing strategy for the element segregation area is as follows: The power is increased proportionally according to the amplitude by which the element concentration exceeds the threshold.

8. The intelligent blanking method for steel structure parts based on laser cutting according to claim 7, wherein: The way of predicting the coupling area in advance is as follows: Extract the boundary coordinates of the risk area from the cutting state matrix, determine the extension direction of the risk zone, and calculate the spatial distribution trend of the risk area; Calculate the time lead for the laser focus to reach the risk area. And during the scanning stage, bind the focus compensation amount and power compensation amount of the risk area to the coordinates and store them in the feedforward control cache queue; When the cutting head enters the prediction time window, read the control parameters in the cache in advance and trigger the actuator to act.

9. A blanking system applying the intelligent blanking method for steel structure parts based on laser cutting according to any one of claims 1-8, characterized in that: It includes: A data acquisition module, which is used to acquire the three-dimensional topography data, reflectivity data, alloy element concentration and alloy element concentration distribution on the surface of the boom main board. A data processing module, which is used to process the acquired data and fuse and construct a real-time cutting state matrix. A region division module, which divides the region type based on the real-time cutting state matrix, and determines the risk region level and its processing strategy of the region type. A control execution module, which executes the corresponding processing strategy based on the risk region level of the region type.

Images