Laser cutting groove quality adaptive control method and system based on visual feedback

Abstract

The application discloses a laser cutting groove quality self-adaptive control method and system based on visual feedback, the method acquires cutting area images in real time through a visual acquisition unit and carries out pretreatment; quality characteristics of three dimensions of angle deviation, slag adhesion degree and slag hanging degree are extracted; after normalization of the characteristics, the characteristics are fused into a comprehensive quality deviation index according to preset weights; the comprehensive deviation is compared with a target threshold to generate a control error signal and carry out amplitude limiting processing; laser power, cutting speed and auxiliary gas pressure are cooperatively adjusted based on the error signal, wherein the laser power adopts an incremental PID control law, the cutting speed introduces a power-speed coupling compensation term to maintain heat input balance, and the auxiliary gas pressure is proportionally adjusted according to the slag adhesion degree; the above steps are repeated until the comprehensive deviation converges below the threshold. The application effectively solves the problems of manual detection lag and difficulty in coordinating parameter coupling, and significantly improves the first-time pass rate and processing consistency.

Description

Technical Field

[0001] This invention belongs to the field of industrial vision and intelligent manufacturing technology, specifically relating to a method and system for adaptive control of laser cutting bevel quality based on visual feedback. Background Technology

[0002] Beveling is a critical process in steel structure manufacturing that determines welding quality. The angle accuracy, surface cleanliness, and edge quality of the bevel directly affect the penetration, weld strength, and overall structural safety of subsequent welds. In shipbuilding, bridge construction, heavy machinery, and other fields, substandard beveling quality will lead to welding defects, thereby affecting the service performance of the entire structural component. Currently, laser cutting or plasma cutting equipment is widely used in the industry for beveling. Laser cutting, due to its advantages of narrow kerf, small heat-affected zone, and high cutting precision, is increasingly widely used in beveling thick plates. However, the quality stability during beveling still faces many challenges. During the cutting process, nonlinear factors such as uneven plate material, thickness fluctuations, cumulative thermal deformation, and fluctuations in auxiliary gas pressure can cause dynamic deviations between the actual bevel angle, slag adhesion, and slag length and the ideal values. If these deviations are not corrected in time, they will directly lead to a decrease in the first-pass yield of the bevel, increase rework costs, and waste of materials and energy.

[0003] Existing methods for bevel quality inspection and control can be mainly divided into three categories. The first category is manual sampling inspection, where operators use tools such as angle gauges and feeler gauges to randomly inspect the bevel angle and surface quality after cutting. If defects are found, parameters such as laser power, cutting speed, or auxiliary gas pressure are manually adjusted based on experience. This method has significant lag; inspection is performed after cutting and cannot be corrected in real time, leading to a large number of defective products being produced in the same batch before they are discovered. Furthermore, manual parameter adjustment is highly dependent on operator experience, resulting in significant differences in results among different personnel, making consistency difficult to guarantee. The second category is offline inspection, using a coordinate measuring machine or dedicated bevel inspection equipment to accurately measure the bevel after cutting. This method has high accuracy but a long inspection cycle, failing to meet the needs of online real-time control, and its high cost makes it difficult to promote in mass production scenarios. The third category is visual inspection, where industrial cameras are installed in the cutting area to capture bevel images for defect identification and classification. However, existing methods mostly remain at the detection-alarm level, i.e., issuing an alarm signal to the operator after identifying a defect, requiring manual parameter adjustment based on the alarm information, and have not yet formed a direct driving mechanism based on the inspection results.

[0004] Furthermore, existing visual inspection methods often focus on identifying single defect types, making it difficult to comprehensively evaluate bevel quality. The strong coupling between laser power, cutting speed, and auxiliary gas pressure also makes it challenging to achieve ideal results through manual adjustments. Therefore, there is an urgent need to provide a technical solution that overcomes these shortcomings. Summary of the Invention

[0005] To address this issue, the present invention provides a method and system for adaptive control of laser cutting bevel quality based on visual feedback, which solves the problems of existing methods that rely on manual post-processing inspection and manual parameter adjustment for bevel cutting quality, resulting in detection lag, difficulty in coordinating parameter coupling, and low first-pass yield.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for laser cutting bevel quality based on visual feedback, comprising the following steps:

[0007] S1. The cutting area image is acquired in real time by the vision acquisition unit, and the cutting area image is preprocessed to obtain a binarized image including the bevel cutting area.

[0008] S2. Based on the binarized image, extract the quality features that reflect the cutting quality, such as the angle deviation, the degree of molten slag adhesion, and the degree of slag buildup. These quality features correspond to the thermal deformation state, the slag removal efficiency state, and the molten pool cooling state during the cutting process, respectively.

[0009] S3. Normalize the quality features and integrate the normalized set dimension features into a comprehensive quality deviation index according to preset weights. Use the comprehensive quality deviation index to uniformly quantify the overall deviation between the current cutting quality and the target quality.

[0010] S4. Compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting processing on the control error signal;

[0011] S5. Based on the control error signal, the control parameters of laser power, cutting speed and auxiliary gas pressure are adjusted in a coordinated manner. The laser power is adjusted by an incremental PID control law. The cutting speed is adjusted by introducing a coupling compensation term on the basis of power adjustment to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion.

[0012] S6. The adjusted control parameters are sent to the actuator in real time for execution, and the image is re-acquired in the next sampling cycle. Steps S1 to S5 are repeated until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality.

[0013] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, step S1, the step of preprocessing the image of the cutting area, includes:

[0014] The pixel values ​​of the red, green, and blue channels of the original color image are multiplied by preset brightness weighting coefficients and then summed to obtain a grayscale image, wherein the sum of the brightness weighting coefficients is 1.

[0015] The grayscale image is subjected to a two-dimensional Gaussian filter, wherein the kernel function of the two-dimensional Gaussian filter is determined by the filter standard deviation, and the filter standard deviation controls the weight distribution at a specified pixel position within the filter window;

[0016] An adaptive threshold segmentation algorithm is used to binarize the filtered image. The adaptive threshold segmentation algorithm aims to maximize the inter-class variance between the foreground and background regions. It searches within the grayscale range to determine the optimal segmentation threshold. The sum of the foreground pixel ratio and the background pixel ratio is 1. The average grayscale value of the foreground and the average grayscale value of the background are obtained by averaging the grayscale values ​​of the pixels in the corresponding regions.

[0017] The filtered image is mapped to a binary image based on the optimal segmentation threshold. In the binary image, the pixel regions with a value of 1 are the bevel cutting regions, and the pixel regions with a value of 0 are the background.

[0018] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, step S2, which involves extracting the angular deviation reflecting the cutting quality, includes:

[0019] Edge detection operators are used to extract edges from the binarized image to obtain the contour edges of the bevel sidewall;

[0020] A Hough transform is performed on the contour edge to detect the straight lines corresponding to the edges on both sides of the bevel. The detected straight lines are represented by a general equation. The two components of the direction vector in the general equation are used to characterize the direction of the straight line, thus avoiding the problem of infinite slope for the vertical edge when the straight line is represented by the slope.

[0021] Calculate the cross product modulus and absolute value of the dot product of the direction vectors of the two lines based on the direction vector components of the general equation of the two edge lines.

[0022] The angle is obtained by using the two-parameter arctangent function to calculate the quotient of the cross product modulus and the absolute value of the dot product. The two-parameter arctangent function determines the quadrant of the angle according to the signs of the numerator and denominator. When the denominator is zero, it returns to 90 degrees. The output angle range is 0 degrees to 90 degrees.

[0023] The absolute angle deviation is obtained by subtracting the measured bevel angle from the target bevel angle, where the target bevel angle is the ideal bevel angle specified in the process drawing.

[0024] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, in step S2, the degree of slag adhesion is characterized by a slag adhesion index, and the extraction step of the slag adhesion index includes:

[0025] In the binarized image, the slag connected domains attached to the bevel sidewall region are identified using a connected domain analysis algorithm;

[0026] The pixel area of ​​the slag connected region is counted as the slag area, and the total pixel area of ​​the bevel cutting region is counted as the bevel area.

[0027] The slag adhesion index is obtained by dividing the slag area by the total bevel area and multiplying the resulting ratio by 100%. The slag adhesion index is expressed as a percentage representing the proportion of slag covering the bevel sidewall surface.

[0028] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, in step S2, the degree of slag adhesion is characterized by the average slag adhesion index, and the extraction step of the average slag adhesion index includes:

[0029] In the binarized image, a number of sampling points are uniformly selected at a preset interval along the bottom edge of the bevel, and the number of sampling points is a preset value;

[0030] For each sampling point, scan outward pixel by pixel along the normal direction of the bottom edge of the bevel, and count the number of consecutive pixels that exceed the ideal contour edge;

[0031] The number of consecutive pixels is converted into physical length based on the pre-calibrated pixel equivalent to obtain the slag length of each sampling point;

[0032] The arithmetic mean of the slag lengths at all sampling points is used to obtain the average slag index, which is expressed in millimeters to represent the average slag length.

[0033] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, step S3, the construction step of the comprehensive quality deviation index includes:

[0034] Divide the absolute angle deviation by the preset maximum permissible angle deviation to obtain the normalized angle deviation term. The absolute angle deviation and the maximum permissible angle deviation have the same angle unit.

[0035] Divide the slag adhesion index by the preset maximum allowable slag adhesion index to obtain the normalized slag adhesion item. The slag adhesion index and the maximum allowable slag adhesion index have the same percentage unit.

[0036] Divide the average slag index by the preset maximum allowable slag length to obtain the normalized slag term. The average slag index and the maximum allowable slag length have the same length unit.

[0037] The normalized angle deviation term, the normalized slag adhesion term, and the normalized slag coating term are multiplied by their respective weighting coefficients and then summed to obtain the comprehensive quality deviation index.

[0038] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, step S4 involves generating a control error signal based on the comparison results and performing amplitude limiting processing on the control error signal, specifically including:

[0039] The original control error is obtained by subtracting the preset target comprehensive deviation threshold from the comprehensive quality deviation index calculated at the current moment.

[0040] Determine whether the absolute value of the original control error is greater than the preset maximum allowable control error;

[0041] If the absolute value of the original control error is greater than the maximum permissible control error, the original control error is clamped to the maximum permissible control error.

[0042] If the absolute value of the original control error is not greater than the maximum permissible control error, the original control error shall remain unchanged.

[0043] After amplitude limiting, a control error signal is obtained, which is a dimensionless quantity.

[0044] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, step S5, the incremental PID control of the laser power includes:

[0045] Obtain the control error at the current sampling time, the control error at the previous sampling time, and the control errors at the previous two sampling times;

[0046] Subtracting the control error of the previous sampling time from the control error of the current sampling time yields the first-order error difference.

[0047] Subtract twice the control error of the previous sampling time from the control error of the current sampling time, and then add the control errors of the previous two sampling times to obtain the second-order difference of the error.

[0048] Multiplying the first-order difference of the error by the power proportional gain yields the proportional control quantity, which is in watts and is used to determine the instantaneous response magnitude to the control error.

[0049] Multiply the control error at the current sampling moment by the power integral gain to obtain the integral control quantity. The power integral gain is in watts and is used to eliminate steady-state error.

[0050] Multiplying the second-order difference of the error by the power differential gain yields the differential control quantity, where the power differential gain is in watts and is used to suppress overshoot and oscillation.

[0051] The laser power increment is obtained by adding the proportional control quantity, integral control quantity, and derivative control quantity.

[0052] The laser power at the previous moment is added to the laser power increment to obtain the laser power at the current moment, and the laser power at the current moment is limited between the laser's rated minimum power and rated maximum power.

[0053] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, in step S5, the cutting speed is adjusted by introducing a coupling compensation term to maintain thermal input balance, specifically including:

[0054] Obtain the control error at the current sampling moment and the laser power increment at the current moment;

[0055] The control error at the current sampling moment is multiplied by the velocity proportional gain to obtain the velocity response term. The velocity proportional gain is in millimeters per second and is used to provide a rapid velocity-direction response to the quality deviation.

[0056] Divide the laser power increment at the current moment by the plate thickness to obtain the power change per unit thickness, where the plate thickness is the measured thickness of the steel plate to be cut.

[0057] Multiply the power change per unit thickness by the power-velocity coupling coefficient to obtain the velocity compensation term. The power-velocity coupling coefficient is obtained through offline calibration experiments. Specifically, a unit power step is applied at a set plate thickness, the velocity change required to maintain stable melt width is measured, and the average slope is taken after fitting the curve.

[0058] Subtracting the speed compensation term from the speed response term yields the cutting speed increment;

[0059] The cutting speed at the previous moment is added to the cutting speed increment to obtain the cutting speed at the current moment, and the cutting speed at the current moment is limited to between the minimum speed and the maximum speed of the servo motor.

[0060] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, the proportional adjustment of the auxiliary gas pressure includes:

[0061] Obtain the slag adhesion index at the current sampling time and the preset target slag index;

[0062] Subtracting the target slag index from the slag adhesion index yields the slag deviation;

[0063] The pressure adjustment amount is obtained by multiplying the slag deviation by the pressure-slag proportional gain. The pressure-slag proportional gain is expressed in megapascals / percentage and is used to map the deviation of the slag adhesion degree into the adjustment range of the auxiliary gas pressure.

[0064] The pressure adjustment amount is added to the preset reference cutting pressure to obtain the auxiliary gas pressure at the current moment. The reference cutting pressure is preset according to the cutting nozzle diameter and the plate thickness.

[0065] As a preferred embodiment of the adaptive control method for laser cutting bevel quality based on visual feedback, in step S6, the convergence condition of the comprehensive quality deviation index is:

[0066] The overall quality deviation index calculated at the current moment is less than the preset convergence threshold, and both the overall quality deviation index and the convergence threshold are dimensionless quantities;

[0067] When the convergence condition is met, the current laser power, cutting speed and auxiliary gas pressure are kept constant, and a stable cutting state is entered.

[0068] When the overall quality deviation index is greater than or equal to the convergence threshold, the image acquisition, feature extraction, overall deviation calculation and parameter adjustment steps are re-executed.

[0069] This invention also provides a visual feedback-based adaptive control system for laser cutting bevel quality, employing the aforementioned visual feedback-based adaptive control method for laser cutting bevel quality, comprising:

[0070] The visual perception module is used to acquire images of the cutting area in real time through the visual acquisition unit, and to preprocess the images of the cutting area to obtain a binarized image including the bevel cutting area.

[0071] The feature extraction module is used to extract quality features reflecting the cutting quality, such as angular deviation, slag adhesion, and slag adhesion, based on the binarized image. These quality features correspond to the thermal deformation state, slag removal efficiency state, and molten pool cooling state during the cutting process, respectively.

[0072] The quality assessment module is used to normalize the quality features and integrate the normalized set dimension features into a comprehensive quality deviation index according to preset weights. The comprehensive quality deviation index is used to uniformly quantify the overall deviation between the current cutting quality and the target quality.

[0073] The error processing module is used to compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting processing on the control error signal.

[0074] An adaptive control module is used to coordinately adjust the control parameters of laser power, cutting speed and auxiliary gas pressure based on the control error signal. The laser power is adjusted using an incremental PID control law. The cutting speed is adjusted by introducing a coupling compensation term on the basis of power adjustment to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion.

[0075] The control parameter distribution module is used to distribute the adjusted control parameters to the actuator in real time and re-acquire the image in the next sampling cycle until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality.

[0076] The present invention has the following advantages:

[0077] This invention acquires images of the cutting area in real time through a visual acquisition unit. After feature extraction and quality assessment, it directly drives the adjustment of cutting parameters. The time interval between detection and correction is shortened to less than a second, completely eliminating the lag of traditional manual inspection and effectively avoiding the generation of batches of defective products.

[0078] This invention normalizes and integrates the three dimensions of angle deviation, slag adhesion index and average slag adhesion index into a single comprehensive quality deviation index, which solves the problem that physical quantities of different dimensions cannot be directly compared and weighted, making the quality evaluation results unique and reproducible, and also providing a unified and clear control basis for subsequent closed-loop control.

[0079] This invention addresses the strong coupling relationship between laser power, cutting speed, and auxiliary gas pressure by constructing a power-speed coupling compensation model. This model maps power increments to speed increments to maintain heat input balance. Simultaneously, it adjusts the auxiliary gas pressure proportionally according to the degree of slag adhesion. The three parameters are coordinated and updated synchronously, overcoming the problems of overshoot, oscillation, or even burn-through that can easily occur when adjusting a single parameter independently. This significantly improves the stability of the thick plate cutting process.

[0080] This invention effectively reduces defects such as bevel angle deviation, slag adhesion, and slag buildup caused by parameter mismatch through a closed-loop control mechanism of real-time detection and automatic correction. This significantly improves the consistency and stability of bevel processing quality, reduces the frequency of rework and grinding, and saves materials and energy consumption.

[0081] The control method of this invention, through the recalibration of preset weighting coefficients and gain parameters, can adapt to the beveling requirements of various materials such as carbon steel, stainless steel, and aluminum alloy, as well as different thickness ranges, and is not limited to specific laser equipment or sheet material. The system can be seamlessly integrated into existing MES systems to achieve intelligent management and data traceability of the beveling process, which is in line with the digital transformation direction of the high-end equipment manufacturing industry. Attached Figure Description

[0082] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0083] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0084] Figure 1 This is a schematic diagram of the adaptive control method for laser cutting bevel quality based on visual feedback provided in an embodiment of the present invention;

[0085] Figure 2 This is a flowchart of the feature extraction process for the adaptive control method of laser cutting bevel quality based on visual feedback provided in this embodiment of the invention.

[0086] Figure 3 The adaptive decision-making flowchart of the laser cutting bevel quality adaptive control method based on visual feedback provided in the embodiments of the present invention is shown below.

[0087] Figure 4 This is an application design architecture diagram of the laser cutting bevel quality adaptive control method based on visual feedback provided in the embodiments of the present invention;

[0088] Figure 5 This is a comparison chart of adaptability to different plate thicknesses provided in the embodiments of the present invention; wherein, (a) bevel angle deviation; (b) slag adhesion index; (c) average slag length; (d) first pass rate;

[0089] Figure 6 This is a comparison chart of the dynamic disturbance suppression effect provided in the embodiments of the present invention; wherein, (a) the peak value of the comprehensive deviation after disturbance; and (b) the time for quality to recover and stabilize.

[0090] Figure 7 This is a batch production consistency analysis diagram provided in an embodiment of the present invention;

[0091] Figure 8 This is a diagram of the architecture of a laser cutting bevel quality adaptive control system based on visual feedback provided in an embodiment of the present invention. Detailed Implementation

[0092] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0093] Example 1

[0094] See Figure 1 , Figure 2 and Figure 3 This invention provides an adaptive control method for laser cutting bevel quality based on visual feedback, comprising the following steps:

[0095] S1. The visual acquisition unit acquires images of the cutting area in real time and preprocesses these images to obtain a binarized image including the bevel cutting area. The original color image acquired by the visual acquisition unit is converted into a computer-processable binarized image that clearly distinguishes the bevel cutting area from the background. The visual acquisition unit is installed behind the cutting head and maintains a fixed angle with the cutting nozzle. It uses an external trigger synchronous mode to acquire images, meaning it acquires images synchronously when the laser cutting pulse is triggered, to avoid interference from high-frequency plasma arc light on image quality. The sampling period is set to 50 milliseconds, i.e., the control frequency is 20 Hz. The preprocessing includes grayscale conversion, Gaussian filtering for noise reduction, and adaptive threshold segmentation. The specific principles of each step are explained in detail in subsequent embodiments.

[0096] S2. Based on the binarized image, extract quality features reflecting the cutting quality: angle deviation, slag adhesion, and slag buildup. These quality features correspond to the thermal deformation state, slag removal efficiency, and molten pool cooling state during the cutting process, respectively. The binarized image obtained in step S1 is converted into quantifiable physical quality features. Only by converting image information into numerical physical quantities can the control algorithm have a clear control object. Angle deviation reflects the bevel angle error caused by thermal deformation of the plate or optical path deviation; slag adhesion index reflects the slag removal efficiency of the auxiliary gas; and average slag buildup index reflects the degree of edge burrs during the cooling and solidification process of the molten pool. These three features are independent yet interconnected, together forming a complete description of the bevel cutting quality, providing a data foundation for the comprehensive evaluation in step S3.

[0097] S3. The quality features are normalized, and the normalized features of the set dimensions are fused into a comprehensive quality deviation index according to preset weights. This comprehensive quality deviation index quantifies the overall deviation between the current cutting quality and the target quality. The quality features of different dimensions extracted in step S2 (angle in degrees, slag in percentage, and slag in millimeters) are normalized to eliminate dimensional differences and then fused into a single dimensionless comprehensive index according to preset weights. If physical quantities of different dimensions are directly weighted and added, the feature with the largest dimension will dominate the comprehensive index, rendering the weight coefficients meaningless. Through normalization, each dimension feature is converted into a relative deviation within the range of 0 to 1, and then weighted and fused to obtain a uniquely determined comprehensive quality deviation index.

[0098] S4. Compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting on the control error signal. The comprehensive quality deviation index calculated in step S3 is converted into an error signal that can be used by the control law, and the error signal is subjected to safety amplitude limiting. The original control error signal is obtained by subtracting the comprehensive deviation from the target threshold (set to 0 in this embodiment). This error signal is dimensionless, and a positive value indicates that the current quality does not meet the standard. The technical necessity of amplitude limiting is that when a serious abnormality occurs during the cutting process (such as a sudden change in the material of the sheet metal, interruption of auxiliary gas, etc.), the comprehensive deviation index may increase sharply. If the error is not limited, the integral term of the subsequent incremental PID will continue to accumulate, causing the control quantity to exceed the rated range of the actuator by a large margin, resulting in system oscillation or even burn-out. By using symmetrical saturation amplitude limiting, it can be ensured that the error signal is always within a reasonable range that the control law can handle, thus ensuring the stability and robustness of the system.

[0099] S5. Based on the control error signal, the control parameters of laser power, cutting speed, and auxiliary gas pressure are adjusted collaboratively. The laser power is regulated using an incremental PID control law. The cutting speed, based on power adjustment, incorporates a coupling compensation term to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion. Based on the control error signal output in step S4, the cutting parameters are adjusted collaboratively. The laser power is regulated using an incremental PID control law, which achieves precise control of heat input by rapidly responding to errors with a proportional term, eliminating steady-state deviations with an integral term, and suppressing overshoot oscillations with a derivative term. The cutting speed, based on power adjustment, incorporates a coupling compensation term to synchronously adjust the speed when the power changes, maintaining a constant heat input per unit length and preventing overheating or burn-through of the molten pool due to increased power. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion, achieving direct control of slag removal efficiency. The adjustment amounts of the three parameters are calculated synchronously and executed within the same sampling period to ensure consistency.

[0100] S6. The adjusted control parameters are sent to the actuator in real time for execution, and the image is reacquired in the next sampling cycle. Steps S1 to S5 are repeated until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality. The parameter adjustment amount calculated in step S5 is sent to the actuator, triggering the next iteration. The adjustment values ​​of laser power, cutting speed and auxiliary gas pressure are sent to the fiber laser, servo motor and gas proportional valve through the real-time industrial bus. The communication cycle is usually 1 millisecond to ensure the real-time transmission of instructions. At the same time, the vision acquisition unit is re-triggered to execute step S1 in the next sampling cycle to form a complete iterative cycle. The convergence condition is set as the comprehensive quality deviation index being less than the preset threshold (0.5 in this embodiment). When the convergence condition is met, the system locks the current parameters and enters a stable cutting state; when the comprehensive deviation increases again, the system immediately reactivates the control algorithm. The typical response time is 0.5 to 2 seconds, which is much faster than the several minutes required for manual detection and correction.

[0101] In one possible embodiment, step S1, the preprocessing step of the image of the cut region, includes:

[0102] The pixel values ​​of the red, green, and blue channels of the original color image are multiplied by preset brightness weighting coefficients and then summed to obtain a grayscale image. The sum of the brightness weighting coefficients is 1. In the original color image, the red, green, and blue channels contribute differently to brightness. The human eye is most sensitive to green and least sensitive to blue; therefore, different brightness weighting coefficients are assigned to each channel according to the characteristics of human vision. In this embodiment, the brightness weighting coefficients are 0.299, 0.587, and 0.114, and their sum is 1. The mathematical expression is as follows:

[0103]

[0104] In the formula, , , The red, green, and blue channels are represented by their pixel coordinates. The pixel value at that location ranges from 0 to 255; The values ​​represent the pixel values ​​of the converted grayscale image, ranging from 0 to 255. The physical significance of this conversion lies in compressing a three-channel image containing rich color information into a single-channel grayscale image, significantly reducing the computational load of subsequent processing while preserving the bevel contour and edge features. Simultaneously, since the plasma arc light causes the most severe interference to the blue channel during the cutting process, this weighted conversion can effectively suppress the interference of arc flicker on edge detection.

[0105] A two-dimensional Gaussian filter is applied to the grayscale image. The kernel function of the two-dimensional Gaussian filter is determined by the filter standard deviation, which controls the weight distribution at specified pixel positions within the filter window. Gaussian filtering is a linear smoothing filter suitable for eliminating Gaussian noise. The two-dimensional Gaussian kernel function is defined as follows:

[0106]

[0107] In the formula, The standard deviation of the Gaussian filter is 1.2 in this embodiment; This represents the pixel coordinate offset relative to the center of the filtering window, in pixels. The filtered image is... , where * is a two-dimensional convolution operator. The value of determines the spatial distribution of the filter kernel function: The larger the value, the more evenly the kernel function distribution, the wider the range of pixels involved in the calculation, the stronger the smoothing and noise reduction effect, but the more severe the edge blurring. The smaller the kernel size, the more concentrated the kernel function distribution, resulting in better edge preservation but weaker denoising capability. This embodiment uses... While effectively removing sensor thermal noise and isolated noise caused by cutting splatter particles, it retains the sharpness of the bevel edge to the greatest extent.

[0108] An adaptive thresholding segmentation algorithm is used to binarize the filtered image. This algorithm aims to maximize the inter-class variance between the foreground and background regions. It iterates through the grayscale range to determine the optimal segmentation threshold, where the sum of the foreground and background pixel proportions is 1. The average grayscale values ​​of the foreground and background are obtained by averaging the grayscale values ​​of the pixels within their respective regions. Based on the image's grayscale distribution characteristics, an optimal threshold is calculated that maximizes the inter-class variance between the foreground and background classes after segmentation. Its mathematical expression is:

[0109]

[0110] In the formula, The candidate segmentation threshold has a value range of 0 to 255. For grayscale values ​​less than or equal to the threshold The percentage of pixels in the background out of the total number of pixels; For grayscale values ​​greater than the threshold The proportion of pixels in the foreground to the total number of pixels, and satisfying the following conditions: ; This represents the average grayscale value of the background area. This represents the average grayscale value of the foreground region.

[0111] The filtered image is mapped to a binary image based on the optimal segmentation threshold. In the binary image, pixel regions with a value of 1 represent the bevel cutting area, and pixel regions with a value of 0 represent the background. The grayscale value of each pixel in the filtered image is then compared with the optimal segmentation threshold. Comparing them one by one, the grayscale value is greater than The pixel is assigned a value of 1, and the grayscale value is no greater than 1. The pixels are assigned a value of 0, thus obtaining a binarized image. .in Represents the bevel cutting area. Background information: This algorithm eliminates the need for manual threshold setting and can adaptively adapt to differences in the reflectivity of different board surfaces, avoiding threshold failure caused by changes in board material or fluctuations in lighting.

[0112] In one possible embodiment, step S2, the step of extracting the angular deviation reflecting the cutting quality, includes:

[0113] Edge detection operators are used to extract edges from the binarized image to obtain the contour edges of the bevel sidewall. Edge detection operators extract the target contour by detecting pixel locations where grayscale values ​​change drastically in the image. In this embodiment, the Canny edge detection operator is used to process the binarized image. The Canny operator is based on gradient calculation. First, it smooths the image using a Gaussian filter, then calculates the gradient magnitude and direction of each pixel, extracts edge pixels through non-maximum suppression and double threshold detection, and finally outputs the contour edges of the bevel sidewall. Compared to operators such as Sobel and Prewitt, the Canny operator has better noise resistance and edge localization accuracy, and can accurately extract the contour information of the bevel sidewall.

[0114] A Hough transform is performed on the contour edge to detect the straight lines corresponding to the edges on both sides of the bevel. The detected straight lines are represented by a general equation. In this general equation, the two components of the direction vector are used to characterize the direction of the straight line, avoiding the problem of infinite slopes for vertical edges when representing the straight line by slope. This is a conventional slope-intercept equation for straight lines. Slope when the straight line is nearly perpendicular Approaching infinity, this can cause numerical overflow in computer floating-point operations. This embodiment uses the general equation for a straight line to represent the detected bevel edge line, namely:

[0115]

[0116]

[0117] In the formula, , , and , , These are the coefficients of the general form equation for the straight lines on the left and right edges of the bevel, respectively. The general form equation contains... It is the normal vector component of the direction vector of the line. The direction of the line can be completely represented by only two components, without having to consider the case of infinite slope. Therefore, it has numerical stability for bevel edges in any direction.

[0118] Calculate the magnitude of the cross product and the absolute value of the dot product of the direction vectors of the two lines based on the direction vector components of the general equation of the two edge lines. The angle between the two lines can be determined by the magnitude of the cross product and the absolute value of the dot product of their direction vectors. For lines... and Their direction vectors are respectively and Cross product modulus The area of ​​the parallelogram formed by two directional vectors is proportional to the sine of the angle between the two vectors; the absolute value of the dot product. The ratio of the projections of two vectors onto each other is proportional to the cosine of the angle between them. The angle can be obtained by using the arctangent function.

[0119] The measured bevel angle is obtained by using a two-parameter arctangent function to calculate the quotient of the absolute value of the cross product modulus and the dot product. The two-parameter arctangent function determines the quadrant of the angle based on the signs of the numerator and denominator, returning 90 degrees when the denominator is zero. The output angle range is 0 to 90 degrees. The formula for calculating the measured bevel angle is as follows:

[0120]

[0121] In the formula, The modulus of the cross product of two direction vectors, physically corresponding to The numerator term; It is the absolute value of the dot product of two directional vectors, and its physical meaning corresponds to The denominator term. It is a two-parameter arctangent function, which differs from a single-parameter function. The difference is: The function automatically determines the quadrant of the angle based on the signs of the numerator and denominator, and avoids division by zero when the denominator is zero (i.e., the two lines are perpendicular), instead directly returning 90 degrees. The output angle range is 0 to 90 degrees, in degrees. This formula applies to two lines in any direction, including cases where one line is perpendicular, thus eliminating the singularity problem inherent in slope representation.

[0122] The absolute angle deviation is obtained by subtracting the measured bevel angle from the target bevel angle, where the target bevel angle is the ideal bevel angle specified in the process drawing. The absolute angle deviation reflects the degree of deviation between the actual bevel angle and the ideal process angle caused by thermal deformation or optical path misalignment during cutting. The calculation formula is as follows:

[0123]

[0124] In the formula, The target bevel angle is the ideal bevel angle specified in the process drawing, and the unit is degrees. The measured bevel angle is in degrees. This represents the absolute angular deviation, expressed in degrees. The larger the deviation value, the more severe the deformation of the sheet metal or the optical path deviation caused by heat accumulation during the cutting process.

[0125] In one possible embodiment, in step S2, the degree of slag adhesion is characterized by a slag adhesion index, and the extraction step of the slag adhesion index includes:

[0126] In the binarized image, connected component analysis algorithms are used to identify the slag connected components attached to the bevel sidewall region. In the binarized image, the slag region and the bevel body region are spatially adjacent but separable connected components. The connected component analysis algorithm scans the binarized image, grouping adjacent pixels with a value of 1 into the same connected component, thereby achieving independent identification and extraction of the slag region. This method is independent of the specific shape or size of the slag and can adapt to the detection needs of slag with different shapes.

[0127] The pixel area of ​​each connected slag region is counted as the slag area, and the total pixel area of ​​the bevel cutting region is counted as the bevel area. The pixel area of ​​the slag region is obtained by counting each pixel of the identified connected slag regions. Count all pixels in the bevel cutting area to obtain the total pixel area of ​​the bevel cutting area. Both areas are expressed in square pixels, based on the same image coordinate system and pixel resolution, to ensure comparability between them.

[0128] The slag adhesion index is obtained by dividing the slag area by the total bevel area and multiplying the resulting ratio by 100%. The slag adhesion index is expressed as a percentage, representing the proportion of slag covering the bevel sidewall surface. The formula for calculating the slag adhesion index is as follows:

[0129]

[0130] In the formula, This represents the pixel area of ​​the slag region, in pixels². This represents the total pixel area of ​​the bevel cutting region, expressed in pixel squares. The slag adhesion index, expressed as a percentage, ranges from 0% to 100%. This index visually reflects the proportion of slag covering the bevel sidewall surface. If the value exceeds the target value, it indicates that the current auxiliary gas pressure is insufficient or the slag adhesion is too strong, and the slag cannot be effectively blown away by the airflow. It is necessary to increase the auxiliary gas pressure or appropriately reduce the cutting speed to increase the slag removal time.

[0131] In one possible embodiment, in step S2, the degree of slag adhesion is characterized by an average slag adhesion index, and the extraction step of the average slag adhesion index includes:

[0132] In the binarized image, a number of sampling points are uniformly selected at preset intervals along the bottom edge of the bevel. The number of sampling points is a preset value. The slag deposits at the bottom of the bevel are unevenly distributed along the edge. If only a single location is selected for detection, the measurement result may not represent the overall slag deposit condition due to randomness. In this embodiment, a number of sampling points are uniformly selected at preset intervals along the bottom edge of the bevel, eliminating the random error of single-point detection through multi-point sampling.

[0133] For each sampling point, scan outward pixel by pixel along the normal direction of the bottom edge of the bevel, and count the number of consecutive pixels exceeding the ideal contour edge. Ideally, the bottom edge of the bevel should lie on a preset straight line. In actual cutting, after the molten pool solidifies, slag may form at the bottom of the bevel that exceeds the ideal contour. For each sampling point, scan outward pixel by pixel along the normal direction of the bottom edge of the bevel (perpendicular to the ideal contour direction), and count the number of consecutive pixels exceeding the ideal contour edge. This number is the pixel length of the slag at that sampling point.

[0134] The number of consecutive pixels is converted into physical length based on a pre-calibrated pixel equivalent to obtain the slag length at each sampling point. Through pre-calibration, a correspondence is established between pixel distances in the image and actual physical distances, i.e., the pixel equivalent (mm / pixel) is calibrated. Multiplying the number of consecutive pixels counted at each sampling point by the pixel equivalent yields the actual slag length at that sampling point. The unit is millimeters.

[0135] The arithmetic mean of the slag lengths at all sampling points is used to obtain the average slag index, which is expressed in millimeters as the average slag length. The formula for calculating the average slag index is as follows:

[0136]

[0137] In the formula, For the first The actual slag length at each sampling point, in millimeters, is converted from pixel distance to physical length using a pre-calibrated pixel equivalent (millimeters / pixel). The total number of sampling points is uniformly distributed along the bottom edge of the bevel; in this embodiment, it is set to 100. The average slag length, expressed in millimeters, reflects the degree of edge burrs caused by insufficient cooling of the molten pool. This indicator, through multi-point sampling, effectively avoids the random errors of single-point detection and can more comprehensively reflect the overall condition of the slag at the bottom of the bevel. If the pressure is too high, it indicates that the cutting speed is too fast, and the molten pool has solidified before it can be blown away by the gas flow. In this case, the cutting speed should be reduced or the auxiliary gas pressure should be increased.

[0138] In one possible embodiment, step S3, the construction step of the comprehensive quality deviation index includes:

[0139] Dividing the absolute angular deviation by a preset maximum permissible angular deviation yields a normalized angular deviation term, where the absolute angular deviation and the maximum permissible angular deviation have the same angular unit. The formula for calculating the normalized angular deviation term is as follows:

[0140]

[0141] In the formula, This refers to the absolute angular deviation, expressed in degrees. The maximum permissible angular deviation is expressed in degrees; in this embodiment, it is taken as 5 degrees. The normalized value is 1 when the angle deviation is 5 degrees. This term is dimensionless and ranges from 0 to 1 (when the angle deviation does not exceed the maximum allowable value). The physical meaning of normalization is to convert the angle deviation into a relative degree of deviation with respect to the maximum tolerance deviation, eliminating the influence of its absolute value on the comprehensive index.

[0142] Dividing the slag adhesion index by a preset maximum allowable slag adhesion index yields the normalized slag adhesion term. The slag adhesion index and the maximum allowable slag adhesion index have the same percentage unit. The formula for calculating the normalized slag adhesion term is as follows:

[0143]

[0144] In the formula, The slag adhesion index is expressed as a percentage. The maximum permissible slag adhesion index, expressed as a percentage, is 20% in this embodiment, i.e., when When the slag adhesion index equals 20%, the normalized value is 1. This term is dimensionless and ranges from 0 to 1 (when the slag adhesion index does not exceed the maximum allowable value). The physical meaning of normalization is to convert the degree of slag adhesion into a relative deviation from the maximum tolerance.

[0145] Dividing the average slag adhesion index by the preset maximum allowable slag adhesion length yields the normalized slag adhesion term. The average slag adhesion index and the maximum allowable slag adhesion length share the same unit of length. The formula for calculating the normalized slag adhesion term is as follows:

[0146]

[0147] In the formula, The average slag-coating length is expressed in millimeters. The maximum allowable slag length is expressed in millimeters; in this embodiment, it is taken as 5 millimeters. The normalized value is 1 when the length is 5 mm. This term is dimensionless and ranges from 0 to 1 (when the average slag length does not exceed the maximum allowable value). The physical meaning of normalization is to convert the degree of slag adhesion into a relative deviation from the maximum allowable length.

[0148] The comprehensive quality deviation index is obtained by multiplying the normalized angle deviation term, the normalized slag adhesion term, and the normalized slag coating term by their respective weighting coefficients and then summing them. The formula for calculating the comprehensive quality deviation index is as follows:

[0149]

[0150] In the formula, , , The weighting coefficients are dimensionless and satisfy the following conditions: In this embodiment, the values ​​are 0.4, 0.3, and 0.3 respectively. Among them, the angle deviation has the most significant impact on the welding quality, so it is given the largest weight. It is a dimensionless comprehensive quality deviation index. A higher value indicates a worse current cutting quality. This indicates that the quality requirements are fully met. The three characteristics—angle deviation (in degrees), slag adhesion index (in percentage), and average slag adhesion index (in millimeters)—have different physical dimensions. If they are directly weighted and added together, the characteristic with the largest dimension will dominate the comprehensive index, rendering the weighting coefficients meaningless. Therefore, this embodiment uses a normalization method, dividing each characteristic by its corresponding maximum allowable value to convert it into a dimensionless relative deviation. The normalization value for each dimension ranges from 0 to 1 (taking a value of 1 when the deviation of that dimension reaches the maximum allowable value). The normalized features are then multiplied by a preset weighting coefficient for weighted fusion, ensuring that the influence weight of each dimension is determined by its relative tolerance upper limit and will not be affected by different dimensions.

[0151] In one possible embodiment, step S4 involves generating a control error signal based on the comparison result and then performing amplitude limiting processing on the control error signal, specifically including:

[0152] The original control error is obtained by subtracting the preset target overall deviation threshold from the overall quality deviation index calculated at the current moment. The calculation formula is:

[0153]

[0154] In the formula, The overall quality deviation index is calculated at the current moment and is dimensionless. The target comprehensive deviation threshold is set to 0 in this embodiment, aiming for zero defects. When... This indicates that the current quality does not meet the standard and control actions need to be taken; when This indicates that the current quality meets the requirements.

[0155] Determine whether the absolute value of the original control error is greater than a preset maximum permissible control error. The mathematical expression of the determination condition is as follows:

[0156]

[0157] In the formula, To determine the maximum permissible control error, this embodiment uses 2.0, meaning the maximum tolerance for the overall deviation is twice the ideal value.

[0158] If the absolute value of the original control error is greater than the maximum permissible control error, the original control error is clamped to the maximum permissible control error. season This involves limiting the amplitude of the error signal to the maximum allowable range. The necessity of this technique lies in the fact that when serious anomalies occur during the cutting process, such as sudden changes in the material of the sheet metal or interruption of the auxiliary gas, the comprehensive deviation index may increase sharply. If the error is not limited, the integral term of the incremental PID will continue to accumulate, causing the control quantity to exceed the rated range of the actuator significantly, leading to system oscillation or even burn-out accidents.

[0159] If the absolute value of the original control error is not greater than the maximum permissible control error, the original control error remains unchanged. season That is, the error signal is used directly as the control error signal without any processing.

[0160] The control error signal is obtained after amplitude limiting, and the control error signal is a dimensionless quantity. Specifically, the control error signal obtained after amplitude limiting... The mathematical expression is:

[0161]

[0162] in, For a symmetric saturated function, all errors are limited to Within the specified range. Symmetrical saturation limiting ensures the error signal remains within a reasonable range manageable by the control law, guaranteeing system stability and robustness. This control error signal... It is a dimensionless quantity, used for power PID control and speed coupling compensation in step S5.

[0163] In one possible embodiment, step S5, the incremental PID control of the laser power includes:

[0164] The control error at the current sampling time, the control error at the previous sampling time, and the control errors at the two previous sampling times are obtained. Incremental PID control requires the error values ​​at the current time and the two previous sampling times to calculate the control increment; therefore, historical error data needs to be stored. In this embodiment, a circular buffer of length 3 is set up inside the controller, and the buffer content is updated with each sampling to store the data. , and Three error values.

[0165] The first-order error difference is obtained by subtracting the control error from the control error at the previous sampling time from the control error at the current sampling time. The formula for calculating the first-order error difference is as follows:

[0166]

[0167] This first-order difference reflects the trend of error change over time, i.e., the rate of change of error, and is used as the input for the proportional control term.

[0168] The second-order error difference is obtained by subtracting twice the control error of the previous sampling time from the control error at the current sampling time, and then adding the control errors of the previous two sampling times. The formula for calculating the second-order error difference is as follows:

[0169]

[0170] This second-order difference reflects the increasing or decreasing trend of the error rate of change. It is used as the input of the differential control term and can detect the trend of the error accelerating in advance, so as to apply damping effect in advance.

[0171] Multiplying the first-order difference of the error by the power proportional gain yields the proportional control quantity, where the power proportional gain is in watts and is used to determine the instantaneous response magnitude to the control error. The formula for calculating the proportional control quantity is as follows:

[0172]

[0173] In the formula, The proportional gain of the power loop, set to 0.5 watts in this embodiment, is used to determine the instantaneous response magnitude to control errors. The proportional term's function is to quickly generate control action to reduce deviation when an error occurs.

[0174] The control error at the current sampling moment is multiplied by the power integral gain to obtain the integral control quantity. The power integral gain, measured in watts, is used to eliminate steady-state error. The formula for calculating the integral control quantity is as follows:

[0175]

[0176] In the formula, The integral gain of the power loop is set to 0.1 watts in this embodiment to eliminate the steady-state error of the system and ensure that the overall quality deviation can converge to near the target value. Since this embodiment uses incremental PID, the integral term is essentially a direct application of the current error, avoiding the integral saturation problem caused by the accumulation of historical errors in positional PID.

[0177] Multiplying the second-order difference of the error by the power differential gain yields the differential control quantity. The power differential gain, measured in watts, is used to suppress overshoot and oscillation. The formula for calculating the differential control quantity is as follows:

[0178]

[0179] In the formula, The differential gain of the power loop, set to 0.2 watts in this embodiment, is used to suppress overshoot and oscillation during the control process, thereby improving the dynamic stability of the system. The differential term anticipates the error change trend based on the second-order difference of the error and applies reverse damping before the error reaches its peak value.

[0180] The proportional control quantity, integral control quantity, and derivative control quantity are added together to obtain the laser power increment. The formula for calculating the laser power increment is as follows:

[0181]

[0182] In the formula, , , These represent the control errors at the current sampling time, the previous sampling time, and the two previous sampling times, respectively; all are dimensionless quantities. The gains mentioned above are all within the sampling period. The offline tuning is set to 50 milliseconds. If the sampling period is changed, it needs to be adjusted according to... Proportional to , Inversely proportional to The relationship must be recalculated; otherwise, the integral or derivative action will be distorted. The advantages of using incremental PID instead of positional PID are: the output of incremental PID is the control increment, which does not depend on historical accumulated errors and will not cause integral saturation problems. At the same time, each output is only related to the error of the three most recent sampling times, resulting in less computation, better real-time performance, and easier implementation of bumpless switching. During the cutting process, when the control parameters need to be adjusted online according to changes in operating conditions, the switching process will not impact the cutting process.

[0183] The laser power at the previous moment is added to the laser power increment to obtain the laser power at the current moment, where the current laser power is limited between the laser's rated minimum power and rated maximum power. The formula for calculating the laser power at the current moment is:

[0184]

[0185] In the formula, The laser power at the current moment is expressed in watts, and is limited by the laser's rated range. In this embodiment, the rated range is 0 to 6000 watts. If the calculated... If the signal exceeds the rated range, it will be clamped to the nearest boundary value to prevent the laser from overloading. When the control error signal... When the result is positive, meaning the quality does not meet the standards. When the value is positive, the laser power increases to improve the heat input; at the same time, the differential term dampens the rate of change of error, suppressing the sharp change in power and avoiding burn-through accidents caused by excessively fast response.

[0186] In one possible embodiment, in step S5, the cutting speed is adjusted by introducing a coupling compensation term to maintain thermal input balance, specifically including:

[0187] The control error at the current sampling moment and the laser power increment at the current moment are obtained. The coupling compensation adjustment of the cutting speed requires simultaneous use of the control error at the current sampling moment. And the laser power increment just calculated in step S5 The control error originates from the output of step S4, while the laser power increment originates from the output of the power PID calculation in step S5. Both belong to the same sampling period in time.

[0188] The control error at the current sampling moment is multiplied by the velocity proportional gain to obtain the velocity response term. This velocity proportional gain, measured in millimeters per second, is used for a rapid velocity-direction response to quality deviations. The formula for calculating the velocity response term is:

[0189]

[0190] In the formula, The speed proportional gain is set to 0.8 mm / s in this embodiment, responsible for a rapid speed-direction response to quality deviations. When the overall quality deviation is positive (quality not up to standard), the speed response term is positive, i.e., the cutting speed is increased.

[0191] Dividing the current laser power increment by the plate thickness yields the power change per unit thickness, where the plate thickness is the measured thickness of the steel plate to be cut. The formula for calculating the power change per unit thickness is:

[0192]

[0193] In the formula, This represents the plate thickness in millimeters, and is the measured thickness of the steel plate to be cut. Physically, this quantity represents the change in power per unit thickness of the plate, reflecting the severity of the impact of power change on the temperature field. Thicker plates have higher heat capacity, so the same change in power has a relatively smaller impact on the temperature field, resulting in a smaller change in power per unit thickness. Conversely, thinner plates have lower heat capacity, resulting in a larger change in power per unit thickness.

[0194] Multiplying the power change per unit thickness by the power-velocity coupling coefficient yields the velocity compensation term. This power-velocity coupling coefficient is obtained through offline calibration experiments. Specifically, a unit power step is applied at a set plate thickness, the velocity change required to maintain stable weld width is measured, and the average slope of the fitted curve is taken. The formula for calculating the velocity compensation term is as follows:

[0195]

[0196] In the formula, The power-velocity coupling coefficient, measured in millimeters squared per (watt-second), represents the velocity compensation required per millimeter of plate thickness for every 1 watt of increased power. In this embodiment, a typical value of 0.05 millimeters squared per (watt-second) is used. This coefficient was obtained through offline calibration experiments, specifically for different plate thicknesses. A unit power step is applied, and the speed change required to maintain stable weld width is measured. The average slope of the fitted curve is then taken. The physical meaning of the power-speed coupling coefficient is: the speed compensation required to maintain stable weld width under unit power change and unit plate thickness conditions.

[0197] Subtracting the speed compensation term from the speed response term yields the cutting speed increment. The formula for calculating the cutting speed increment is:

[0198]

[0199] During laser cutting, the heat input per unit length is When laser power increases If the cutting speed remains constant, an increase in heat input per unit length can lead to quality problems such as overheating of the molten pool and burn-through of the bevel. Therefore, the cutting speed must be increased simultaneously with the increase in power to maintain a constant heat input per unit length and ensure the stability of the cutting process. The thickness of the sheet metal... The technical effect of introducing it as a denominator is that thick plates have a large heat capacity, so the same change in power has a relatively small impact on the temperature field of the thick plate, therefore the speed compensation should be reduced accordingly; conversely, thin plates have a small heat capacity, so the change in power is more sensitive to the temperature field, therefore the speed compensation should be increased accordingly. As the denominator, it perfectly achieves this adaptive adjustment effect.

[0200] The cutting speed at the previous moment is added to the cutting speed increment to obtain the cutting speed at the current moment. The current cutting speed is limited between the minimum and maximum speed of the servo motor. The formula for calculating the cutting speed at the current moment is:

[0201]

[0202] In the formula, The cutting speed at the current moment is expressed in millimeters per second. Limited by the maximum speed of the servo motor, the adjustable speed range in this embodiment is 0 to 100 millimeters per second. If the calculated... If the value exceeds the adjustable range, it will be clamped to the nearest boundary value to ensure that the servo motor operates within a safe operating range.

[0203] In one possible embodiment, the proportional adjustment of the auxiliary gas pressure includes:

[0204] Obtain the slag adhesion index at the current sampling time and the preset target slag index. Slag adhesion index The quality features extracted in step S2 are calculated at the current sampling time; the target slag index. This is a system preset value, allowing for trace amounts of adhesion to ensure the surface quality of the bevel.

[0205] The slag deviation is obtained by subtracting the target slag index from the slag adhesion index. The formula for calculating the slag deviation is as follows:

[0206]

[0207] In the formula, This represents the slag adhesion index at the current sampling time, expressed as a percentage. The target slag index is expressed as a percentage. This represents the slag deviation, expressed as a percentage, indicating the amount of deviation between the current degree of slag adhesion and the target value.

[0208] Multiplying the slag deviation by the pressure-slag proportional gain yields the pressure adjustment amount. This pressure-slag proportional gain, expressed in megapascals per percentage, maps the deviation in slag adhesion to an adjustment in the auxiliary gas pressure. The formula for calculating the pressure adjustment amount is as follows:

[0209]

[0210] In the formula, This is the pressure-slag proportional gain, expressed in megapascals per percentage, used to map deviations in slag adhesion to adjustments in the auxiliary gas pressure. When the slag deviation... When the value is positive (i.e., the slag adhesion exceeds the target value), the pressure adjustment is positive, that is, the auxiliary gas pressure is increased to enhance the purging capacity.

[0211] The pressure adjustment amount is added to a preset reference cutting pressure to obtain the current auxiliary gas pressure. The reference cutting pressure is preset based on the cutting nozzle diameter and the plate thickness. The formula for calculating the current auxiliary gas pressure is:

[0212]

[0213] In the formula, The pressure of the auxiliary gas at the current moment is expressed in megapascals (MPA). The reference cutting pressure, measured in megapascals (MPa), is preset based on the cutting nozzle diameter and plate thickness. The technical principle behind this proportional adjustment is that the primary function of the auxiliary gas is to blow away the molten slag generated during the cutting process from the kerf. When the slag adhesion index... Exceeding the target value If the current air pressure is insufficient to completely remove the molten slag, then... The slag adhesion index will increase to enhance the blowing capacity; conversely, when the slag adhesion index decreases... Below the target value If the pressure is too high, it indicates that the current gas pressure is too high. Under the premise of ensuring the slag removal effect, the gas pressure should be appropriately reduced to save gas energy consumption.

[0214] In one possible embodiment, in step S6, the convergence condition of the comprehensive quality deviation index is:

[0215] The calculated overall quality deviation index at the current moment is less than a preset convergence threshold, where both the overall quality deviation index and the convergence threshold are dimensionless quantities. The mathematical expression of the convergence condition is as follows:

[0216]

[0217] In the formula, The overall quality deviation index is calculated at the current moment and is dimensionless. The convergence threshold represents the point at which the overall quality deviation has been reduced to within the acceptable range.

[0218] because and Both are dimensionless quantities, and can be directly compared without unit conversion. When the convergence condition is met, the current laser power, cutting speed, and auxiliary gas pressure remain unchanged, entering a stable cutting state. When the convergence condition is met, the system determines that the current cutting quality has met the standard, and maintains the current laser power, cutting speed, and auxiliary gas pressure unchanged, entering a stable cutting state. At this time, the control algorithm pauses parameter adjustment to avoid frequent fluctuations in control parameters due to minor noise, thus improving the steady-state performance of the system.

[0219] When the overall quality deviation index is greater than or equal to the convergence threshold, the image acquisition, feature extraction, overall deviation calculation, and parameter adjustment steps are re-executed. When the overall quality deviation index is greater than or equal to the convergence threshold, the system determines that the current cutting quality does not meet the standard and re-executes the image acquisition, feature extraction, overall deviation calculation, and parameter adjustment steps, forming a continuously iterative closed-loop control process. The typical response time is 0.5 to 2 seconds, depending on the severity of the defect, which is much faster than the several minutes required for manual inspection and correction.

[0220] See Figure 4This paper presents the system application design architecture of a visual feedback-based adaptive control method for laser cutting bevel quality. The architecture adopts a four-layer progressive processing structure, consisting of a bottom-level perception layer, a feature extraction layer, a quality assessment layer, and an adaptive decision layer from bottom to top. The physical execution of control commands is completed through the bottom actuator layer, and finally, the entire system is connected into a continuously iterative closed-loop control loop through visual feedback.

[0221] The bottom perception layer is responsible for real-time image acquisition during the cutting process. An industrial camera continuously captures images of the cutting area at a resolution of 1920×1080 and a frame rate of 30fps. A narrowband filter, with its center wavelength matched to the laser wavelength, suppresses strong light interference such as arcing and splashing, ensuring image quality meets subsequent analysis requirements. An external trigger synchronization module uses encoder signals to achieve precise synchronization between image acquisition and the cutting position, ensuring that each frame corresponds to a specific location on the cutting path. After this hardware collaboration, the system outputs the original grayscale image of the cutting area and transmits it to the feature extraction layer.

[0222] The feature extraction layer is responsible for processing the input image step by step, extracting three core feature parameters that can quantify the cutting quality. First, the grayscale conversion module performs weighted average grayscale conversion on the image to reduce data dimensionality while preserving edge structure information. Subsequently, the Gaussian filtering module further smooths and denoises, suppressing high-frequency noise while protecting edge integrity. Based on this, Otsu adaptive threshold binarization segments the image into foreground and background regions, providing a clear binarization foundation for subsequent feature calculations. Above this, the architecture performs three feature calculations in parallel: first, it extracts precise contours through Canny edge detection, fits the straight lines on both sides of the bevel using Hough transform, and calculates the included angle to obtain the bevel angle; second, it uses connected component analysis to mark the slag region and calculates its area ratio to obtain the slag adhesion index; third, it calculates the average distance exceeding the ideal contour along the normal direction of the bottom edge of the cut to obtain the average slag adhesion index. After the three feature parameters are completed, they are passed to the quality evaluation layer.

[0223] The quality assessment layer transforms the three heterogeneous features from the feature extraction layer into a single quality index usable for decision-making. Normalization eliminates the dimensional differences between the bevel angle, slag adhesion index, and average slag adhesion index, mapping each feature to a unified interval. Subsequently, a weighted fusion calculation multiplies each normalized feature by its corresponding weight coefficient and sums the results. These weight coefficients are statistically set based on historical quality data. The final output comprehensive deviation function provides the adaptive decision layer with a quantifiable quality deviation signal.

[0224] The adaptive decision layer generates multi-channel coordinated control commands based on the comprehensive deviation signal. The incremental PID controller calculates the laser power adjustment based on the deviation value, achieving precise closed-loop power regulation through the coordinated use of proportional, integral, and derivative terms. The power-speed coupling compensation model synchronously adjusts the cutting speed according to power changes, ensuring a constant energy input per unit length and preventing a decrease in cutting quality due to power fluctuations. The gas pressure proportional controller dynamically adjusts the auxiliary gas pressure based on the slag adhesion index, increasing the pressure to enhance the slag removal effect when there is excessive slag. The three control signals are output collaboratively to form a complete control command, which is then sent to the actuator layer.

[0225] The actuator layer comprises three types of actuators. A fiber laser receives power commands and adjusts its output power in real time within the range of 0 to 6000W; a servo motor, in conjunction with a guide rail, drives the cutting head to move at a set speed within the range of 0 to 100 mm / s; and a gas proportional valve proportionally adjusts the flow and pressure of the auxiliary gas within the range of 0 to 2.5 MPa. These three types of actuators work together to physically control the cutting process. After control execution, the system returns to visual feedback, and the underlying perception layer re-acquires the image of the cutting area in the next acquisition cycle. This forms a complete closed-loop iteration from image acquisition, feature extraction, quality assessment to adaptive decision-making and execution, until the cutting quality meets the preset standard.

[0226] To verify the effectiveness of the proposed adaptive control method for laser cutting bevel quality based on visual feedback, the specific experimental verification scheme and results are presented below.

[0227] I. Experimental Objective

[0228] The method of the present invention was verified to have an adaptive control effect on bevel angle deviation, slag adhesion and slag defects during thick plate bevel cutting. It was compared with the traditional manual parameter adjustment method to prove the advantages of the present invention in terms of bevel pass rate, cutting quality consistency and response speed.

[0229] II. Test Conditions

[0230] The experiment used a fiber laser cutting machine with a rated laser power range of 0 to 6000 watts and an adjustable cutting speed range of 0 to 100 mm / s. Compressed air was used as the auxiliary gas, with an adjustable pressure range of 0 to 2.5 MPa. The vision acquisition system employed an industrial camera with a resolution of 1920×1080, a frame rate of 30 frames / second, and a sampling period of 50 milliseconds. It was mounted on the side and rear of the cutting head, maintaining an angle of approximately 35 degrees with the cutting nozzle, and equipped with a narrow-band filter to eliminate arc light interference.

[0231] The test plates are marine steel plate DH36, with three specifications of 10 mm, 16 mm and 25 mm in thickness respectively. The size of each test piece is 300 mm × 200 mm, and the target bevel angle is set to 30 degrees. A control group and a test group are set for each plate specification for comparison. The control group adopts the traditional manual parameter adjustment method, that is, during the cutting process, the operator manually sets the laser power, cutting speed and auxiliary gas pressure based on experience, and conducts spot inspection and adjusts the parameters after cutting is completed; the test group adopts the method of the present invention for online closed-loop adaptive control.

[0232] III. Test Scheme

[0233] The test is divided into three groups, each group carries out a comparative test, and the specific scheme is as follows.

[0234] The first group is the adaptability test for different plate thicknesses. Bevel cutting is performed on plates with three thicknesses of 10 mm, 16 mm and 25 mm respectively, 20 test pieces are cut in each group, of which the first 10 are cut in the manner of the control group and the last 10 are cut in the manner of the test group. During the cutting process, the bevel angle deviation, slag adhesion index and average slag hanging index of each group of test pieces are recorded, and the one-pass acceptance rate is counted. The acceptance criteria for the one-pass acceptance rate are: the bevel angle deviation does not exceed plus or minus 2 degrees of the target angle, the slag adhesion index does not exceed 20%, and the average slag hanging length does not exceed 5 mm. A test piece is deemed qualified only if it meets all three conditions at the same time.

[0235] The second group is the dynamic disturbance suppression test. A material mutation region (such as a slag inclusion or abnormal hardness section) is preset in the middle of the plate to simulate the non-uniform working condition disturbance during the cutting process. For each plate thickness specification, 5 control group cutting operations and 5 test group cutting operations are performed respectively, and the maximum value of the comprehensive quality deviation index during the cutting process and the time required from the occurrence of the disturbance to the recovery of stable quality are recorded.

[0236] The third group is the batch consistency test. A plate with a thickness of 16 mm is used to continuously cut 30 test pieces, all using the test group method. The mean value and standard deviation of the bevel angle deviation, slag adhesion index and average slag hanging length of the 30 test pieces are counted to verify the ability of the method of the present invention to guarantee the consistency of batch processing quality.

[0237] IV. Test Results

[0238] (1) Results of the adaptability test for different plate thicknesses

[0239] Under the conditions of three different plate thicknesses, the comparison of cutting quality between the control group and the test group is shown in Table 1 and Figure 5 as shown in the following.

[0240] Table 1 Comparison of cutting quality between the control group and the test group under different plate thickness conditions

[0241]

[0242] from Figure 5 As shown in Table 1, the first-pass yield of the control group decreased significantly with increasing plate thickness, reaching only 30% for a 25mm thick plate. In contrast, the experimental group achieved a 100% first-pass yield for 10mm and 16mm thick plates, and a 90% first-pass yield for a 25mm thick plate. The experimental group showed significantly better performance than the control group in three indicators: angle deviation, slag adhesion index, and average slag length. Particularly for the 25mm thick plate, the angle deviation decreased from ±4.1 degrees to ±1.3 degrees, and the slag adhesion index decreased from 26% to 8%, demonstrating a significant improvement.

[0243] Experimental results show that the method of the present invention can effectively control the bevel cutting quality under different plate thicknesses, especially for 25 mm thick plates, where the first-pass yield is significantly improved compared to manual methods. Meanwhile, as the plate thickness increases, the difficulty and error rate of manual parameter adjustment increase significantly, while the method of the present invention maintains stable cutting quality through adaptive compensation of plate thickness parameters.

[0244] (II) Results of Dynamic Disturbance Suppression Test

[0245] After setting a material abrupt change region in the middle of the board, the performance of the two control methods differed significantly. The comparison of the disturbance responses of the control group and the experimental group is shown in Table 2. Figure 6 As shown.

[0246] Table 2 Comparison of responses between the control group and the experimental group under dynamic disturbance conditions.

[0247]

[0248] In the control group, when cutting to the area of ​​material abrupt change, the cutting status fluctuated significantly, the bevel angle deviation increased instantaneously, and the slag adhesion index rose sharply. Operators usually needed to perform inspections after cutting to detect quality problems, and parameter adjustment time was approximately five to ten minutes. Some specimens had to be scrapped due to failure to adjust parameters in time, resulting in bevel burn-through or severe slag adhesion.

[0249] In the test group, when cutting to a region of material abrupt change, the visual acquisition unit detected an increase in the overall quality deviation index within approximately 100 to 200 milliseconds after the disturbance occurred. The adaptive control unit completed parameter adjustments within 0.8 to 1.6 seconds, causing the overall quality deviation index to fall back below the convergence threshold. No manual intervention was required throughout the process, and no burn-through or serious defects occurred.

[0250] Experimental results show that the method of this invention has the ability to quickly sense external disturbances and adaptively adjust parameters, reducing the response time from minutes to seconds, effectively avoiding the generation of batches of defective products. It should be noted that in the 25 mm thick plate test group, the peak value of the comprehensive deviation after disturbance was 2.0, which is on the edge of the amplitude limiting condition. This indicates that under severe disturbance conditions, the control law is close to full-amplitude output, but the system can still remain stable and successfully converge, verifying the necessity of amplitude limiting processing.

[0251] (III) Results of Batch Consistency Test

[0252] Thirty specimens were continuously cut from a 16 mm thick sheet, and all were tested using the method of this invention. The test results are shown in Table 3 and... Figure 7 As shown.

[0253] Table 3. Statistical results of batch consistency test (30 specimens)

[0254]

[0255] As shown in Table 3, the average angular deviation of the 30 specimens was ±1.1 degrees, with a standard deviation of ±0.3 degrees, which is significantly smaller than the standard deviation of ±3 degrees under manual parameter adjustment. The average slag adhesion index was 6%, with a standard deviation of 1.5%. The average slag adhesion length was 2.0 mm, with a standard deviation of 0.5 mm. The first-pass yield of the 30 specimens was 100%, with no specimens requiring rework.

[0256] It should be noted that the average angle deviation (±1.1 degrees) in Table 3 is consistent with the data of the 16 mm test group in Table 1. The two data points come from different sample sizes (30 specimens and 10 specimens), and the data agreement indicates that the method of this invention has good repeatability and stability. The adjustment of control parameters during the cutting process was also recorded. During the processing of 30 specimens, the laser power was adjusted 8 times, the cutting speed was adjusted 12 times, and the auxiliary gas pressure was adjusted 6 times. All adjustments occurred after the parameters of the first two specimens converged. The parameters of subsequent specimens remained basically stable, and the system achieved continuous and stable cutting without human intervention. Batch consistency test results show that the method of this invention has stable quality control capabilities in continuous processing scenarios, and the quality dispersion between specimens is far lower than that of manual methods.

[0257] V. Experimental Conclusions

[0258] Based on the above three sets of experiments, the following conclusions can be drawn.

[0259] First, the method of this invention effectively improves the first-pass yield of beveling. Under different plate thicknesses, the first-pass yield of specimens using the method of this invention is significantly higher than that of traditional manual parameter adjustment methods, especially for thick plate cutting, where the improvement in first-pass yield is even more significant.

[0260] Second, the method of this invention has the ability to suppress dynamic disturbances. Faced with sudden material inhomogeneity disturbances during the cutting process, the method of this invention can automatically adjust parameters and restore a stable cutting state within seconds, while manual methods require several minutes and are difficult to guarantee in a timely manner.

[0261] Third, the method of this invention ensures the consistency of batch processing quality. The dispersion of angular deviation, slag adhesion, and slag length of 30 consecutively cut specimens is significantly less than that of manual methods, verifying the effect of closed-loop control on improving quality consistency.

[0262] In summary, the method of this invention can effectively solve the problems of detection lag, strong dependence on manual operation, and difficulty in coordinating parameter coupling in the prior art, and realizes online adaptive closed-loop control of bevel cutting quality, which has significant industrial application value.

[0263] Example 2

[0264] See Figure 8 Embodiment 2 of the present invention also provides a laser cutting bevel quality adaptive control system based on visual feedback, which adopts the laser cutting bevel quality adaptive control method based on visual feedback described in the above embodiments, including:

[0265] The visual perception module 100 is used to acquire images of the cutting area in real time through the visual acquisition unit, and to preprocess the images of the cutting area to obtain a binarized image including the bevel cutting area.

[0266] The feature extraction module 200 is used to extract quality features reflecting the cutting quality, such as angular deviation, slag adhesion and slag adhesion, based on the binarized image. The quality features correspond to the thermal deformation state, slag removal efficiency state and molten pool cooling state during the cutting process, respectively.

[0267] The quality assessment module 300 is used to normalize the quality features and integrate the normalized set dimension features into a comprehensive quality deviation index according to preset weights. The comprehensive quality deviation index is used to uniformly quantify the overall deviation between the current cutting quality and the target quality.

[0268] The error processing module 400 is used to compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting processing on the control error signal.

[0269] The adaptive control module 500 is used to coordinately adjust the control parameters of laser power, cutting speed and auxiliary gas pressure based on the control error signal. The laser power is adjusted using an incremental PID control law. The cutting speed is adjusted by introducing a coupling compensation term on the basis of power adjustment to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion.

[0270] The control parameter sending module 600 is used to send the adjusted control parameters to the actuator in real time for execution, and to re-acquire the image in the next sampling cycle until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality.

[0271] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0272] Example 3

[0273] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a laser cutting bevel quality adaptive control method based on visual feedback. The program code includes instructions for executing the laser cutting bevel quality adaptive control method based on visual feedback of Embodiment 1 or any possible implementation thereof.

[0274] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0275] Example 4

[0276] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0277] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the visual feedback-based adaptive control method for laser cutting bevel quality according to Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0278] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0279] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0280] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0281] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for adaptive control of laser cutting bevel quality based on visual feedback, characterized in that, Includes the following steps: S1. The cutting area image is acquired in real time through the vision acquisition unit, and the cutting area image is preprocessed to obtain a binarized image including the bevel cutting area. S2. Based on the binarized image, extract the quality features that reflect the cutting quality, such as the angle deviation, the degree of molten slag adhesion, and the degree of slag buildup. These quality features correspond to the thermal deformation state, the slag removal efficiency state, and the molten pool cooling state during the cutting process, respectively. S3. Normalize the quality features and integrate the normalized set dimension features into a comprehensive quality deviation index according to a preset weight. Use the comprehensive quality deviation index to uniformly quantify the overall deviation between the current cutting quality and the target quality. S4. Compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting processing on the control error signal; S5. Based on the control error signal, the control parameters of laser power, cutting speed and auxiliary gas pressure are adjusted in a coordinated manner. The laser power is adjusted by an incremental PID control law. The cutting speed is adjusted by introducing a coupling compensation term on the basis of power adjustment to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion. S6. The adjusted control parameters are sent to the actuator in real time for execution, and the image is re-acquired in the next sampling cycle. Steps S1 to S5 are repeated until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality.

2. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 1, characterized in that, Step S1, the preprocessing step for the image of the cut region includes: The pixel values ​​of the red, green, and blue channels of the original color image are multiplied by preset brightness weighting coefficients and then summed to obtain a grayscale image, wherein the sum of the brightness weighting coefficients is 1. The grayscale image is subjected to a two-dimensional Gaussian filter, wherein the kernel function of the two-dimensional Gaussian filter is determined by the filter standard deviation, and the filter standard deviation controls the weight distribution at a specified pixel position within the filter window; An adaptive threshold segmentation algorithm is used to binarize the filtered image. The adaptive threshold segmentation algorithm aims to maximize the inter-class variance between the foreground and background regions. It searches within the grayscale range to determine the optimal segmentation threshold. The sum of the foreground pixel ratio and the background pixel ratio is 1. The average grayscale value of the foreground and the average grayscale value of the background are obtained by averaging the grayscale values ​​of the pixels in the corresponding regions. The filtered image is mapped to a binary image based on the optimal segmentation threshold. In the binary image, the pixel regions with a value of 1 are the bevel cutting regions, and the pixel regions with a value of 0 are the background.

3. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 1, characterized in that, Step S2, the step of extracting the angular deviation reflecting the cutting quality includes: Edge detection operators are used to extract edges from the binarized image to obtain the contour edges of the bevel sidewall; A Hough transform is performed on the contour edge to detect the straight lines corresponding to the edges on both sides of the bevel. The detected straight lines are represented by a general equation. The two components of the direction vector in the general equation are used to characterize the direction of the straight line, thus avoiding the problem of infinite slope for the vertical edge when the straight line is represented by the slope. Calculate the cross product modulus and absolute value of the dot product of the direction vectors of the two lines based on the direction vector components of the general equation of the two edge lines. The angle is obtained by using the two-parameter arctangent function to calculate the quotient of the cross product modulus and the absolute value of the dot product. The two-parameter arctangent function determines the quadrant of the angle according to the signs of the numerator and denominator. When the denominator is zero, it returns to 90 degrees. The output angle range is 0 degrees to 90 degrees. The absolute angle deviation is obtained by subtracting the measured bevel angle from the target bevel angle, where the target bevel angle is the ideal bevel angle specified in the process drawing.

4. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 3, characterized in that, In step S2, the degree of slag adhesion is characterized by the slag adhesion index, and the extraction steps of the slag adhesion index include: In the binarized image, the slag connected domains attached to the bevel sidewall region are identified using a connected domain analysis algorithm; The pixel area of ​​the slag connected region is counted as the slag area, and the total pixel area of ​​the bevel cutting region is counted as the bevel area. The slag adhesion index is obtained by dividing the slag area by the total bevel area and multiplying the resulting ratio by 100%. The slag adhesion index is expressed as a percentage of the proportion of slag covering the bevel sidewall surface. In step S2, the degree of slag adhesion is characterized by the average slag adhesion index, and the extraction steps of the average slag adhesion index include: In the binarized image, a number of sampling points are uniformly selected at a preset interval along the bottom edge of the bevel, and the number of sampling points is a preset value; For each sampling point, scan outward pixel by pixel along the normal direction of the bottom edge of the bevel, and count the number of consecutive pixels that exceed the ideal contour edge; The number of consecutive pixels is converted into physical length based on the pre-calibrated pixel equivalent to obtain the slag length of each sampling point; The arithmetic mean of the slag lengths at all sampling points is used to obtain the average slag index, which is expressed in millimeters to represent the average slag length.

5. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 4, characterized in that, In step S3, the steps for constructing the comprehensive quality deviation index include: Divide the absolute angle deviation by the preset maximum permissible angle deviation to obtain the normalized angle deviation term. The absolute angle deviation and the maximum permissible angle deviation have the same angle unit. Divide the slag adhesion index by the preset maximum allowable slag adhesion index to obtain the normalized slag adhesion item. The slag adhesion index and the maximum allowable slag adhesion index have the same percentage unit. Divide the average slag index by the preset maximum allowable slag length to obtain the normalized slag term. The average slag index and the maximum allowable slag length have the same length unit. The normalized angle deviation term, the normalized slag adhesion term, and the normalized slag coating term are multiplied by their respective weighting coefficients and then summed to obtain the comprehensive quality deviation index.

6. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 5, characterized in that, In step S4, a control error signal is generated based on the comparison result, and the control error signal is subjected to amplitude limiting processing, specifically including: The original control error is obtained by subtracting the preset target comprehensive deviation threshold from the comprehensive quality deviation index calculated at the current moment. Determine whether the absolute value of the original control error is greater than the preset maximum allowable control error; If the absolute value of the original control error is greater than the maximum permissible control error, the original control error is clamped to the maximum permissible control error. If the absolute value of the original control error is not greater than the maximum permissible control error, the original control error shall remain unchanged. After amplitude limiting, a control error signal is obtained, which is a dimensionless quantity.

7. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 6, characterized in that, In step S5, the incremental PID control of the laser power includes: Obtain the control error at the current sampling time, the control error at the previous sampling time, and the control errors at the previous two sampling times; Subtracting the control error of the previous sampling time from the control error of the current sampling time yields the first-order error difference. Subtract twice the control error of the previous sampling time from the control error of the current sampling time, and then add the control errors of the previous two sampling times to obtain the second-order difference of the error. Multiplying the first-order difference of the error by the power proportional gain yields the proportional control quantity, which is in watts and is used to determine the instantaneous response magnitude to the control error. Multiply the control error at the current sampling moment by the power integral gain to obtain the integral control quantity. The power integral gain is in watts and is used to eliminate steady-state error. Multiplying the second-order difference of the error by the power differential gain yields the differential control quantity, where the power differential gain is in watts and is used to suppress overshoot and oscillation. The laser power increment is obtained by adding the proportional control quantity, integral control quantity, and derivative control quantity. The laser power at the previous moment is added to the laser power increment to obtain the laser power at the current moment, and the laser power at the current moment is limited between the laser's rated minimum power and rated maximum power.

8. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 7, characterized in that, In step S5, the cutting speed is adjusted by introducing a coupling compensation term to maintain thermal input balance, specifically including: Obtain the control error at the current sampling moment and the laser power increment at the current moment; The control error at the current sampling moment is multiplied by the velocity proportional gain to obtain the velocity response term. The velocity proportional gain is in millimeters per second and is used to provide a rapid response in the velocity direction to the quality deviation. Divide the laser power increment at the current moment by the plate thickness to obtain the power change per unit thickness, where the plate thickness is the measured thickness of the steel plate to be cut. Multiply the power change per unit thickness by the power-velocity coupling coefficient to obtain the velocity compensation term. The power-velocity coupling coefficient is obtained through offline calibration experiments. Specifically, a unit power step is applied at a set plate thickness, the velocity change required to maintain stable melt width is measured, and the average slope is taken after fitting the curve. Subtracting the speed compensation term from the speed response term yields the cutting speed increment; The cutting speed at the previous moment is added to the cutting speed increment to obtain the cutting speed at the current moment, and the cutting speed at the current moment is limited to between the minimum speed and the maximum speed of the servo motor. The proportional adjustment of the auxiliary gas pressure includes: Obtain the slag adhesion index at the current sampling time and the preset target slag index; Subtracting the target slag index from the slag adhesion index yields the slag deviation; The pressure adjustment amount is obtained by multiplying the slag deviation by the pressure-slag proportional gain, which is expressed in megapascals / percentage and is used to map the deviation of the slag adhesion degree into the adjustment range of the auxiliary gas pressure. The pressure adjustment amount is added to the preset reference cutting pressure to obtain the auxiliary gas pressure at the current moment. The reference cutting pressure is preset according to the cutting nozzle diameter and the plate thickness.

9. The adaptive control method for laser cutting bevel quality based on visual feedback according to claim 1, characterized in that, In step S6, the convergence condition for the comprehensive quality deviation index is: The overall quality deviation index calculated at the current moment is less than the preset convergence threshold, and both the overall quality deviation index and the convergence threshold are dimensionless quantities; When the convergence condition is met, the current laser power, cutting speed and auxiliary gas pressure are kept constant, and a stable cutting state is entered. When the overall quality deviation index is greater than or equal to the convergence threshold, the image acquisition, feature extraction, overall deviation calculation and parameter adjustment steps are re-executed.

10. A visual feedback-based adaptive control system for laser cutting bevel quality, employing the visual feedback-based adaptive control method for laser cutting bevel quality as described in any one of claims 1 to 9, characterized in that, include: The visual perception module is used to acquire images of the cutting area in real time through the visual acquisition unit, and to preprocess the images of the cutting area to obtain a binarized image including the bevel cutting area. The feature extraction module is used to extract quality features reflecting the cutting quality, such as angular deviation, slag adhesion, and slag adhesion, based on the binarized image. These quality features correspond to the thermal deformation state, slag removal efficiency state, and molten pool cooling state during the cutting process, respectively. The quality assessment module is used to normalize the quality features and integrate the normalized set dimension features into a comprehensive quality deviation index according to preset weights. The comprehensive quality deviation index is used to uniformly quantify the overall deviation between the current cutting quality and the target quality. The error processing module is used to compare the comprehensive quality deviation index with the target threshold, generate a control error signal based on the comparison result, and perform amplitude limiting processing on the control error signal; An adaptive control module is used to coordinately adjust the control parameters of laser power, cutting speed and auxiliary gas pressure based on the control error signal. The laser power is adjusted using an incremental PID control law. The cutting speed is adjusted by introducing a coupling compensation term on the basis of power adjustment to maintain heat input balance. The auxiliary gas pressure is adjusted proportionally according to the degree of molten slag adhesion. The control parameter distribution module is used to distribute the adjusted control parameters to the actuator in real time and re-acquire the image in the next sampling cycle until the comprehensive quality deviation index converges to below the preset threshold, so as to realize online adaptive control of the bevel cutting quality.

Images