An automatic copper pipe cutting method and system based on visual guidance

Abstract

The present invention discloses a method and system for automatic cutting of copper tubes based on visual guidance, specifically related to the technical field of automatic cutting of copper tubes. It uses a high-frame-rate infrared camera to collect the thermal radiation texture during the automatic cutting of copper tubes, extracts features through wavelet transform and optical flow algorithm, enhances the signal using a generative adversarial network, constructs a multi-scale thermal stress distribution model of copper tubes, generates a thermal stress distribution map in real time and corrects the dynamic threshold, and realizes adaptive cutting in combination with the Bezier curve algorithm; based on the dynamic utilization of thermal radiation texture, multi-scale thermal effect coupling and morphological evolution threshold optimization, it improves the cutting accuracy and environmental adaptability to solve the problem that when the thermal radiation texture generated by local heating during the copper tube cutting process is used as visual guidance information, it is difficult for the visual system to stably extract effective features due to the randomness and instantaneousness of the thermal radiation texture, thus affecting the reliability of the cutting path optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic copper tube cutting, and more specifically, the present invention relates to a method and system for automatic copper tube cutting based on visual guidance. Background Art

[0002] In the industrial production of automatic copper tube cutting, the temperature in the cutting area rapidly rises from room temperature to a local maximum of over 500 °C. An infrared camera can capture the thermal radiation texture generated by local heating. This texture reflects the temperature distribution and thermal stress state of the copper tube in the form of light and dark changes, providing potential visual guidance information for dynamically optimizing the cutting path. Compared with traditional cutting methods that rely on preset marks, using the thermal radiation texture eliminates the need for additional marking steps and can directly reflect real-time process characteristics, helping to avoid material expansion or surface quality deterioration caused by overheating.

[0003] Existing infrared imaging technologies mostly focus on the temperature detection of static targets and lack a systematic processing scheme for the spatio-temporal changes and noise interference of thermal radiation textures in a dynamic cutting environment, restricting their effective application in optimizing the cutting accuracy and reliability of copper tubes.

[0004] When the thermal radiation texture generated by local heating during the copper tube cutting process is used as visual guidance information, the visual system is difficult to stably extract effective features due to the randomness and instantaneousness of the thermal radiation texture, thus affecting the reliability of cutting path optimization. Specifically, during the visual guidance process based on the thermal radiation texture, the randomness, instantaneousness of the texture and its confounding effect with environmental thermal noise interfere with feature extraction. The generation of the thermal radiation texture is affected by factors such as laser power fluctuations, copper tube wall thickness deviations, and auxiliary air flow disturbances, presenting complex dynamic change characteristics. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for automatic copper tube cutting based on visual guidance. Through thermal radiation texture acquisition and feature extraction, signal separation and enhancement, and thermal stress modeling and correction, the adaptive cutting of copper tubes is finally realized to solve the problem that when the thermal radiation texture generated by local heating during the copper tube cutting process is used as visual guidance information, the visual system is difficult to stably extract effective features due to the randomness and instantaneousness of the thermal radiation texture, thus affecting the reliability of cutting path optimization as mentioned in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solution: A method for automatic copper tube cutting based on visual guidance, comprising the following steps:

[0007] Step 1. Thermal radiation texture acquisition and feature extraction: Real-time acquisition of the original thermal radiation texture image of the copper tube cutting area through a high-frame-rate infrared camera, and simultaneous recording of the environmental parameters during the cutting process; using wavelet transform technology to analyze the spatial frequency distribution of the original thermal radiation texture image, marking the areas of thermal stress concentration and the edge positions of high temperature gradients; combining with the optical flow algorithm, calculating the dynamic change vector of the thermal radiation texture, quantifying the evolution trend of the thermal radiation texture over time, and obtaining the thermal radiation texture feature dataset of the thermal radiation texture, including the spatial frequency map, the dynamic change vector field, and the corresponding environmental parameter table;

[0008] Explanation: Ensuring texture consistency through cross-stage correction means using cross-stage correction techniques (such as the dynamic time warping algorithm) to process the thermal radiation texture feature dataset, adjusting the texture morphological differences caused by heat accumulation or diffusion in different cutting stages (such as initial heating, mid-cutting, and final cooling), ensuring the consistency and stability of the feature dataset throughout the cutting process, and thus providing reliable basic data for subsequent signal enhancement and thermal stress modeling;

[0009] Step 2. Signal separation and enhancement: Perform image segmentation and signal enhancement processing on the original thermal radiation texture image. Based on the thermal radiation texture feature dataset, use image segmentation to extract effective thermal radiation texture signals; enhance the clarity and continuity of the thermal radiation texture image through signal enhancement processing, and strengthen the prominence of the thermal stress distribution characteristics near the cutting line; during the image segmentation process, use the spatial frequency map as the weight to distinguish the high-stress areas and low-correlation noise areas in the thermal radiation texture; use a generative adversarial network to enhance clarity and continuity, generate an enhanced image sequence, extract texture features, and mark the confidence score of each texture feature;

[0010] Step 3. Thermal stress modeling and correction: Based on the enhanced image sequence, combined with the material properties of the copper tube, construct a thermal stress distribution model of the copper tube. Use the finite element analysis method to estimate the temperature field and stress field during the cutting process, where the temperature field is calculated through the heat conduction equation, and the stress field is derived based on the strain caused by thermal expansion; generate and update the thermal stress distribution map during the copper tube cutting process in real time, mark the boundary of the cutting guidance area based on the thermal stress distribution map, and adjust the dynamic threshold of the guidance path to ensure the cutting accuracy. Output the real-time updated thermal stress distribution map, and mark the boundary coordinates of the cutting guidance area and the adjustment log of the dynamic threshold;

[0011] Step 4. Realize adaptive cutting of copper tube: Use the Bezier curve algorithm to generate an adaptive cutting path and execute it. Optimize the cutting parameters according to the dynamic threshold adjustment log, and generate a cutting report including path coordinates, thermal radiation texture features, and optimized parameters.

[0012] Preferably, in the above Step 1, it includes the step of cross-stage correction of thermal radiation texture:

[0013] Cluster the time series of the thermal radiation texture feature dataset through the dynamic time warping algorithm, and analyze the evolution trend of the texture morphology in different cutting stages;

[0014] Based on the clustering results, use the feature alignment technology to adjust the dynamically changing vector field, so that the texture features across stages are consistent, generate the corrected thermal radiation texture feature dataset, and transfer it to the signal separation and enhancement steps.

[0015] Preferably, in the first step, an adaptive frame rate adjustment step is included, and the sampling frame rate of the infrared camera is adjusted in real time according to the dynamic change speed of the thermal radiation texture or the environmental parameters during the cutting process to balance the data quality and processing efficiency.

[0016] Preferably, the confidence score S of the texture feature is obtained as follows:

[0017] Obtain the feature instantaneous vector v_t, denote the average vector of the feature in the historical time series as v_avg, and calculate the feature time series consistency index C through the formula where T is the time window; the feature time series consistency index C represents the stability of the texture feature in the time series;

[0018] Extract the probability P of each texture feature in the discriminator, and combine the consistency index C of the feature in the dynamically changing vector field, and calculate the confidence score S through the following formula:

[0019] ;

[0020] where P is the probability value predicted by the discriminator, and α is an adjustment coefficient that controls the influence of the feature time series consistency on the confidence score.

[0021] Preferably, in the process of obtaining the thermal stress distribution map, use the environmental parameters during the cutting process to calibrate the parameters of the copper tube thermal stress distribution model, input the enhanced image sequence as the initial condition into the calibrated copper tube thermal stress distribution model, and output the real-time updated thermal stress distribution map through iterative calculation, reflecting the temperature gradient and stress concentration area on the surface of the copper tube.

[0022] Preferably, the third step includes a multi-scale thermal stress coupling step: based on the enhanced image sequence, use the multi-resolution decomposition technology to separate the microscopic and macroscopic thermal effect features; combine the microscopic defect distribution data of the copper tube material to construct a multi-scale copper tube thermal stress distribution model; use the data fusion algorithm to integrate the microscopic stress field and the macroscopic stress field to generate a coupled thermal stress distribution map, improving the accuracy of the cutting guide area boundary.

[0023] Preferably, the method further includes an air flow noise management step to solve the problem of interference of the thermal noise signal caused by the auxiliary air flow and to distinguish between pseudo-texture and real thermal radiation texture. Monitoring is started in real time during the cutting process. When the auxiliary air flow acts on the surface of the copper tube, the air flow noise management step is triggered, including:

[0024] Step S11, monitoring the auxiliary air flow parameters and predicting the noise characteristics: Recording the air flow parameters of the auxiliary air flow, including the pressure and flow rate parameters of the auxiliary air flow; establishing a mapping relationship between the air flow noise and the image characteristics to obtain a pseudo-texture prediction model, which is used to output the brightness and spectral characteristics of the pseudo-texture;

[0025] Step S12, extracting and analyzing the spatio-temporal characteristics of the texture: Obtaining the infrared image sequence and the pseudo-texture prediction result, using a 3D convolutional neural network to analyze the spatio-temporal characteristics of the infrared image, generating a spatio-temporal feature map, and suppressing the random interference of the pseudo-texture based on the regular characteristics of the real thermal radiation texture;

[0026] Step S13, enhancing the real texture characteristics and marking the confidence level: Processing the texture data in the spatio-temporal feature map, using an attention mechanism to strengthen the brightness and spectral characteristics of the real thermal radiation texture; combining the discriminator probability and the feature temporal consistency to label a confidence score for each texture feature.

[0027] Preferably, the dynamic threshold is obtained in the following way:

[0028] Comparing the enhanced image sequence with the prediction result of the copper tube thermal stress distribution model, and calculating the dynamic threshold of the thermal stress concentration area. The specific method is:

[0029] Extracting the boundary with the largest temperature gradient in the enhanced image, performing spatial matching with the stress concentration area predicted by the copper tube thermal stress distribution model, and calculating the matching degree D;

[0030] Based on the matching result, the dynamic threshold formula is defined as , where is the reference temperature, γ is the gain coefficient, and H is the spatial complexity of the texture feature; adjusting γ according to the texture feature confidence score S, and γ is positively correlated with the confidence score S to ensure that the high-confidence features have a greater impact on the threshold;

[0031] By the time trend of the historical time series enhanced image sequence, using the Kalman filter to correct the instantaneous fluctuation, removing the outliers, and generating a smoothed dynamic threshold sequence; correcting the instantaneous random fluctuation of the thermal radiation texture through time series analysis and removing the outliers.

[0032] Preferably, the matching degree is calculated by the morphological evolution of the thermal stress concentration area, and the calculation method of the dynamic threshold is:

[0033] ;

[0034] ;

[0035] Among them, is the second-order gradient of the image at time t, is the second-order gradient of the image at the initial time, is the probability distribution of each frequency component in the spatial frequency map, and T is the time window.

[0036] To achieve the above object, the present invention provides the following technical solutions: An automatic copper pipe cutting system based on visual guidance, comprising:

[0037] A thermal texture acquisition module that real-time acquires the original thermal radiation texture image of the copper pipe cutting area through a high-frame-rate infrared camera and synchronously records the environmental parameters of the cutting process;

[0038] A feature extraction module that uses wavelet transform technology to analyze the spatial frequency distribution, marks the thermal stress concentration area and the high temperature gradient edge; combines the optical flow algorithm to calculate the dynamic change vector of the thermal radiation texture, quantifies the evolution trend of the thermal radiation texture over time, captures the instantaneous and random features, and generates a thermal radiation texture feature dataset, including a spatial frequency map, a dynamic change vector field, and an environmental parameter table;

[0039] A signal enhancement module that performs image segmentation on the original thermal radiation texture image based on the thermal radiation texture feature dataset, extracts the effective thermal radiation texture signal; enhances the image clarity and continuity through a generative adversarial network, generates an enhanced image sequence, extracts texture features and annotates a confidence score for each feature;

[0040] A stress modeling module that receives the enhanced image sequence, combines the copper pipe material properties, constructs a copper pipe thermal stress distribution model, and real-time generates and updates the thermal stress distribution map; annotates the boundary of the cutting guide area based on the thermal stress distribution map and adjusts the dynamic threshold to ensure the cutting accuracy; outputs the real-time updated thermal stress distribution map, the cutting guide area boundary coordinates, and the dynamic threshold adjustment log;

[0041] A path optimization module that generates an adaptive cutting path based on the thermal stress distribution map, the cutting guide area boundary coordinates, and the dynamic threshold adjustment log using the Bezier curve algorithm; optimizes the cutting parameters in combination with the dynamic threshold adjustment log, performs the cutting operation and generates a cutting report.

[0042] The technical effects and advantages of the present invention:

[0043] The automatic copper tube cutting method based on vision guidance provided by the present invention can effectively detect and identify the thermal stress distribution during the copper tube cutting process by collecting thermal radiation images in real time and using wavelet transform and optical flow algorithm to analyze the dynamic changes of thermal radiation texture in detail; using an enhanced generative adversarial network to process and enhance the thermal radiation texture image to make the thermal stress distribution characteristics near the cutting line more obvious for fine cutting; by constructing a thermal stress distribution model of the copper tube and combining with the optimization and adjustment of the dynamic cutting path, it can dynamically adjust the cutting parameters according to the changes of thermal stress during the actual cutting process to improve the cutting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of the automatic copper tube cutting method of the present invention.

[0045] Figure 2 It is a flowchart of the air flow noise management of the present invention.

[0046] Figure 3 It is a structural block diagram of the automatic copper tube cutting system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0048] At the same time, it should be understood that, for the sake of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship.

[0049] The following description of at least one exemplary embodiment is merely illustrative in nature and in no way serves as a limitation on the present application and its application or use.

[0050] The techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods, and devices should be regarded as part of the specification.

[0051] Example 1, referring to Figure 1 the flowchart of the automatic copper tube cutting method, the present invention provides a Figure 1 kind of automatic copper tube cutting method based on vision guidance as shown in

[0052] Step 1. Thermal radiation texture acquisition and feature extraction: Real-time acquisition of the original thermal radiation texture image of the copper tube cutting area through a high-frame-rate infrared camera, and simultaneous recording of the environmental parameters during the cutting process (such as laser power, cutting speed, and auxiliary air flow pressure); using wavelet transform technology to analyze the spatial frequency distribution of the original thermal radiation texture image, marking the areas of thermal stress concentration and the edge positions of high temperature gradients; combining with the optical flow algorithm, calculating the dynamic change vector of the thermal radiation texture, quantifying the evolution trend of the thermal radiation texture over time, and then capturing the instantaneous and random features of the thermal radiation texture, obtaining the thermal radiation texture feature dataset of the thermal radiation texture, including the spatial frequency map, the dynamic change vector field, and the corresponding environmental parameter table;

[0053] Step 2. Signal separation and enhancement: Perform image segmentation and signal enhancement processing on the original thermal radiation texture image. Based on the thermal radiation texture feature dataset, use image segmentation to extract effective thermal radiation texture signals; enhance the clarity and continuity of the thermal radiation texture image through signal enhancement processing, and strengthen the prominence of the thermal stress distribution characteristics near the cutting line; during the image segmentation process, use the spatial frequency map as the weight to distinguish the high-stress area from the low-correlation noise area in the thermal radiation texture; use a generative adversarial network to enhance clarity and continuity, generate an enhanced image sequence, extract texture features, and mark the confidence score of each texture feature;

[0054] Step 3. Thermal stress modeling and correction: Real-time generate and update the thermal stress distribution map during the copper tube cutting process, mark the boundaries of the cutting guide area based on the thermal stress distribution map, and adjust the dynamic threshold of the guide path to ensure cutting accuracy, output the real-time updated thermal stress distribution map, and mark the boundary coordinates of the cutting guide area and the adjustment log of the dynamic threshold;

[0055] Explanation: Perform cutting guidance based on the dynamic threshold;

[0056] Step 4. Realize adaptive cutting of copper tubes: Use the Bezier curve algorithm to generate and execute an adaptive cutting path, optimize the cutting parameters according to the dynamic threshold adjustment log, and generate a cutting report containing path coordinates, thermal radiation texture features, and optimized parameters.

[0057] Background description: During the copper tube cutting process, the thermal radiation texture changes significantly with the movement of the laser and the accumulation of heat. For example, in the initial stage, the texture is a uniform strip (temperature gradient 100°C / mm), and in the middle and late stages, it becomes irregular patches due to heat diffusion (gradient drops to 50°C / mm). The feature extraction in Step 1 does not design a consistency correction mechanism for this cross-stage morphological evolution, which may lead to subsequent signal enhancement and thermal stress modeling based on inconsistent feature inputs, affecting cutting accuracy. Most existing technologies assume that the texture features are relatively stable and do not pay attention to the dynamic consistency problem. Based on this:

[0058] In a possible embodiment, in the first step, it includes a thermal radiation texture cross-stage correction step:

[0059] Cluster the time series of the thermal radiation texture feature dataset through the dynamic time warping algorithm, and analyze the texture morphology evolution trend in different cutting stages;

[0060] Based on the clustering results, use the feature alignment technology to adjust the dynamically changing vector field, so that the cross-stage texture features are consistent, generate the corrected thermal radiation texture feature dataset, and transfer it to the second step for signal separation and enhancement.

[0061] In the embodiments of the present invention, it needs to be further explained that the first step includes an adaptive frame rate adjustment step, which adjusts the sampling frame rate of the infrared camera in real time according to the dynamic change speed of the thermal radiation texture or the cutting process environment parameters to balance data quality and processing efficiency; it includes:

[0062] Use the optical flow algorithm to calculate the dynamic change vector of the thermal radiation texture and quantify the instantaneous speed of texture change;

[0063] Based on the pre-trained machine learning model, combined with the change trends of different materials and process parameters, predict the frame rate under the current conditions by analyzing historical cutting data and environment parameters;

[0064] According to the quantified texture change speed and the prediction result of the machine learning model, dynamically adjust the sampling frame rate of the infrared camera to ensure optimizing data processing efficiency while capturing instantaneous features.

[0065] In a possible embodiment, adjust the sampling frame rate according to the spatial frequency change rate of the thermal radiation texture and the non-linear interaction of the environment parameters; the non-linear interaction means that the environment parameters do not have a simple linear superposition relationship, but jointly affect the thermal radiation texture through complex physical mechanisms (such as heat conduction, convection, radiation); analyze the non-linear interaction of the environment parameters and predict the impact on the texture frequency through a non-linear regression model.

[0066] In the embodiments of the present invention, it needs to be further explained that the method for obtaining the confidence score S of the texture feature is:

[0067] Obtain the feature instantaneous vector v_t, denote the average vector of the feature in the historical time series (such as the previous 5 frames) as v_avg, and calculate the feature time series consistency index C through the formula where T is the time window (for example, 0.5 seconds); the feature time series consistency index C represents the stability of the texture feature in the time series;

[0068] Extract the probability P of each texture feature in the discriminator, and combine the consistency index C of the feature in the dynamically changing vector field to calculate the confidence score S through the following formula:

[0069] ;

[0070] Among them, P is the probability value predicted by the discriminator, and α is the adjustment coefficient, which controls the influence of the feature temporal consistency on the confidence score (for example, the value range is from 0.1 to 1.0); the formula uses the exponential function to reflect the non-linear influence of the feature temporal consistency on the confidence score, avoiding the simplistic treatment of the traditional linear weighted summation method, making the confidence score have a stronger response ability to the dynamically changing feature points.

[0071] In the embodiments of the present invention, it needs to be further explained that the acquisition method of the thermal stress distribution map is as follows: based on the enhanced image sequence output in step two, combined with the material properties of the copper tube (including thermal conductivity, thermal expansion coefficient, density, specific heat capacity, etc.), a copper tube thermal stress distribution model is constructed, and the finite element analysis method is used to estimate the temperature field and stress field during the cutting process, where the temperature field is calculated by the heat conduction equation, and the stress field is derived based on the strain caused by thermal expansion;

[0072] Calibrate the parameters of the copper tube thermal stress distribution model by using the cutting process environment parameters (such as laser power 2000W, cutting speed 10 mm / s, auxiliary air flow pressure 5 bar) (for example, the heat input power and heat dissipation coefficient in the boundary conditions, the laser power determines the heat input amount, and the air flow pressure affects the heat dissipation rate); input the enhanced image sequence as the initial condition into the calibrated copper tube thermal stress distribution model, and output the real-time updated thermal stress distribution map through iterative calculation, reflecting the temperature gradient and stress concentration area on the surface of the copper tube.

[0073] Background description, the copper tube thermal stress distribution model is based on the enhanced image sequence and material properties, using finite element analysis, but does not consider the multi-scale coupling of micro (such as material grain boundary defects, size 1 - 10 μm) and macro (such as overall heat diffusion, size 1 - 10 mm) thermal effects. For example, micro-defects may amplify local stress (local stress increases to 80 MPa), affecting the accuracy of the macro distribution map. Existing technologies are mostly based on single-scale modeling and do not solve the multi-scale interaction problem. Based on this:

[0074] In a possible embodiment, step three includes a multi-scale thermal stress coupling step: based on the enhanced image sequence, use the multi-resolution decomposition technique (such as pyramid transformation) to separate the micro and macro thermal effect features;

[0075] Combine the copper tube material micro-defect distribution data (pre-obtained by ultrasonic detection) to construct a multi-scale copper tube thermal stress distribution model;

[0076] Integrate the microscopic stress field (local features) and the macroscopic stress field using a data fusion algorithm (such as weighted least squares method) to generate a coupled thermal stress distribution map, improving the accuracy of the boundary of the cutting guidance area.

[0077] In the embodiments of the present invention, it needs to be further explained that the multi-scale thermal stress coupling step includes:

[0078] Microscopic feature extraction: Based on the enhanced image sequence, use wavelet packet decomposition technology to separate the microscopic thermal effect features, extract the local temperature gradient change with a scale less than 10 μm, and combine the microscopic defect distribution data of the copper tube material to generate a microscopic thermal effect feature map;

[0079] Macroscopic feature calculation: Analyze the enhanced image sequence through two-dimensional Fourier transform, extract the macroscopic thermal diffusion pattern with a scale greater than 1 mm, and generate a macroscopic thermal effect feature map;

[0080] Multi-scale fusion: Use an adaptive weighted fusion algorithm to integrate the microscopic thermal effect feature map and the macroscopic thermal effect feature map, and define the fusion weight , where is the microscopic temperature gradient, is the macroscopic temperature gradient, and λ is an adjustment coefficient (value range 0.1 - 0.5);

[0081] Generate a coupled thermal stress distribution map through iterative optimization to ensure that the boundary coordinate accuracy error is less than 0.1 mm.

[0082] In the embodiments of the present invention, it needs to be further explained that the multi-scale thermal stress coupling step includes:

[0083] Multi-scale feature separation: Based on the enhanced image sequence, use variational mode decomposition technology to separate the microscopic thermal effect features (frequency range 10 - 100 Hz) and the macroscopic thermal effect features (frequency range 0.1 - 1 Hz), and generate corresponding microscopic feature maps and macroscopic feature maps;

[0084] Defect influence quantification: Combine the microscopic defect distribution data of the copper tube material, and calculate the microscopic stress correction through the local stress amplification factor , where d is the defect size (μm), D is the wall thickness of the copper tube (mm), is the material-related coefficient (value range 0.5 - 2);

[0085] Dynamic coupling calculation: Use a deep belief network to fuse the microscopic feature map and the macroscopic feature map, train the model to predict the dynamic influence of defects on the thermal stress distribution, and generate a real-time updated coupled thermal stress distribution map; Adjust the boundary of the cutting guidance area according to the prediction result to ensure that the deviation between the boundary coordinates and the actual stress concentration area is less than 0.15 mm.

[0086] In the background description, air refractive index perturbations caused by auxiliary airflows or environmental heat dissipation will superimpose thermal noise signals on the image, which are highly similar to the real thermal radiation texture in terms of brightness and spectral characteristics, leading to misjudgment of the thermally stressed concentration area by the system, and further causing the cutting path to deviate from the target position. To solve the problem that the pseudo-textures caused by auxiliary airflows are difficult to identify and filter out:

[0087] In the embodiments of the present invention, it needs to be further explained that in step two, there is also an air flow noise management step to solve the interference of thermal noise signals caused by auxiliary airflows and distinguish between pseudo-textures and real thermal radiation textures. During the cutting process, monitoring is started in real time. When the auxiliary air flow acts on the surface of the copper tube, the air flow noise management step is triggered. Refer to Figure 2 the air flow noise management flowchart of, including:

[0088] Step S11, monitor the auxiliary air flow parameters and predict the noise characteristics: Record the air flow parameters of the auxiliary air flow, including the pressure and flow rate parameters of the auxiliary air flow; establish the mapping relationship between the air flow noise and the image characteristics to obtain the pseudo-texture prediction model. The pseudo-texture prediction model is used to output the brightness and spectral characteristics of the pseudo-texture as the reference basis for subsequent filtering;

[0089] Step S12, extract and analyze the spatio-temporal characteristics of the texture: Obtain the infrared image sequence and the pseudo-texture prediction result, and use a 3D convolutional neural network to analyze the spatio-temporal characteristics of the infrared image (for example, the real thermal radiation texture should remain consistent in multiple time frames, while the randomness of the air flow noise may fluctuate in time), generate the spatio-temporal feature map, and based on the regular characteristics of the real thermal radiation texture, suppress the random interference of the pseudo-texture;

[0090] Step S13, enhance the real texture characteristics and mark the confidence level: Process the texture data in the spatio-temporal feature map, and use the attention mechanism to strengthen the brightness and spectral characteristics of the real thermal radiation texture; combine the discriminator probability and the feature temporal consistency to label the confidence score for each texture feature;

[0091] Explanation, assuming that at a certain moment, the enhanced thermal radiation texture image shows a strong thermally stressed concentration area; through the Bayesian optimization algorithm, according to the characteristics and confidence scores of this image, the system will dynamically adjust the threshold of the cutting path; if the thermal radiation texture of a certain area is very clear and credible, the threshold will be correspondingly reduced to accurately cut this area; if the thermal radiation texture is relatively fuzzy, the system will increase the threshold to avoid mis-cutting.

[0092] In the embodiments of the present invention, it needs to be further explained that the acquisition method of the dynamic threshold is:

[0093] Compare the enhanced image sequence with the prediction results of the copper tube thermal stress distribution model, and calculate the dynamic threshold of the thermal stress concentration area. The specific method is as follows:

[0094] Extract the boundary with the largest temperature gradient in the enhanced image (such as the area where the gradient ≥ 200 °C / mm), and perform spatial matching with the stress concentration area predicted by the copper tube thermal stress distribution model (such as the stress ≥ 50 MPa), and calculate the matching degree (for example, use cosine similarity to quantify the spatial consistency between the image and the model, and regard the cosine similarity as the matching degree D. When the matching degree is high, close to 1, the enhanced image and the model are highly consistent, and the threshold increases greatly; when the matching degree D is low, the threshold is conservative to reduce misjudgment);

[0095] Based on the matching results, define the dynamic threshold formula as , where is the reference temperature, γ is the gain coefficient, and H is the spatial complexity of the texture feature; adjust γ according to the texture feature confidence score S, and γ is positively correlated with the confidence score S to ensure that high-confidence features have a greater impact on the threshold;

[0096] In a possible embodiment, enhance the time trend of the image sequence through historical time series (such as the first 10 frames), use Kalman filtering to correct the instantaneous fluctuations, and eliminate outliers (such as temperature mutations of ±25 °C) to generate a smoothed dynamic threshold sequence; through time series analysis (based on the texture trend of the first 10 frames of images, using the exponential smoothing method with α = 0.3), correct the instantaneous random fluctuations of the thermal radiation texture, and eliminate outliers to ensure that the dynamic threshold reflects the stable thermal stress distribution characteristics.

[0097] In a possible embodiment, use the morphological evolution of the thermal stress concentration area to represent the matching degree, and the calculation method of the dynamic threshold is as follows:

[0098] ;

[0099] ;

[0100] where is the reference temperature (calculated based on the median of the local temperature of the enhanced image), is the gain coefficient, is the second-order gradient of the image at time t, is the second-order gradient of the image at the initial time, is the spatial complexity of the texture feature, is the probability distribution of each frequency component in the spatial frequency map (similar to Shannon entropy); in this formula, the gradient change amount It represents the degree of change in the texture structure in the image, reflecting the morphological evolution of the thermo-stress concentration area during the cutting process. The integral term captures the evolution of the texture over time, enhancing the response ability of the dynamic threshold to the thermo-stress changes during the cutting process. Combining the time smoothing of the threshold by the Kalman filter further removes outliers, ensuring the stability and accuracy of the dynamic threshold.

[0101] Example 2. Refer to Figure 3 the structural block diagram of the automatic copper tube cutting system. The embodiment of the present invention provides an automatic copper tube cutting system based on vision guidance, including:

[0102] A thermal texture acquisition module that real-time acquires the original thermal radiation texture image of the copper tube cutting area through a high-frame-rate infrared camera, and simultaneously records the environmental parameters during the cutting process (such as laser power, cutting speed, and auxiliary air flow pressure);

[0103] A feature extraction module that uses wavelet transform technology to analyze the spatial frequency distribution, marking the thermo-stress concentration area and the high temperature gradient edge; combining the optical flow algorithm to calculate the dynamic change vector of the thermal radiation texture, quantifying the evolution trend of the thermal radiation texture over time, capturing the instantaneous and random features, and generating a thermal radiation texture feature dataset, including a spatial frequency map, a dynamic change vector field, and an environmental parameter table;

[0104] A signal enhancement module that performs image segmentation on the original thermal radiation texture image based on the thermal radiation texture feature dataset, extracting the effective thermal radiation texture signal; enhancing the image clarity and continuity through a generative adversarial network, generating an enhanced image sequence, extracting texture features and annotating a confidence score for each feature;

[0105] A stress modeling module that receives the enhanced image sequence, combines the copper tube material properties (such as thermal conductivity, thermal expansion coefficient, etc.), constructs a copper tube thermo-stress distribution model, and real-time generates and updates the thermo-stress distribution map; annotates the boundary of the cutting guidance area based on the thermo-stress distribution map and adjusts the dynamic threshold to ensure the cutting accuracy; outputs the real-time updated thermo-stress distribution map, the boundary coordinates of the cutting guidance area, and the dynamic threshold adjustment log;

[0106] A path optimization module that generates an adaptive cutting path using the Bezier curve algorithm based on the thermo-stress distribution map, the boundary coordinates of the cutting guidance area, and the dynamic threshold adjustment log; combines the dynamic threshold adjustment log to optimize the cutting parameters (such as laser power or air flow pressure), performs the cutting operation, and generates a cutting report.

[0107] In the embodiment of the present invention, it needs to be further explained that the thermal texture acquisition module includes an adaptive frame rate adjustment unit that real-time adjusts the sampling frame rate of the infrared camera according to the dynamic change speed of the thermal radiation texture or the environmental parameters during the cutting process to balance the data quality and processing efficiency.

[0108] In the embodiments of the present invention, it needs to be further explained that the feature extraction module includes a thermal radiation texture cross-stage correction unit, which clusters the time series of the thermal radiation texture feature dataset through the dynamic time warping algorithm, and analyzes the texture morphology evolution trend in different cutting stages;

[0109] Based on the clustering results, the feature alignment technology is used to adjust the dynamically changing vector field, so that the cross-stage texture features are kept consistent, and a corrected thermal radiation texture feature dataset is generated and transmitted to the signal enhancement module.

[0110] In the embodiments of the present invention, it needs to be further explained that the signal enhancement module includes an air flow noise management unit to solve the problem of thermal noise signal interference caused by the auxiliary air flow and distinguish between pseudo-textures and real thermal radiation textures. During the cutting process, the monitoring is started in real time. When the auxiliary air flow acts on the surface of the copper pipe, the air flow noise management unit is triggered, including:

[0111] Step S11, monitoring the auxiliary air flow parameters and predicting the noise characteristics: recording the air flow parameters of the auxiliary air flow, including the pressure and flow rate parameters of the auxiliary air flow; establishing the mapping relationship between the air flow noise and the image features to obtain a pseudo-texture prediction model, which is used to output the brightness and spectral characteristics of the pseudo-texture;

[0112] Step S12, extracting and analyzing the spatio-temporal features of the texture: obtaining the infrared image sequence and the pseudo-texture prediction result, and using a 3D convolutional neural network to analyze the spatio-temporal features of the infrared image to generate a spatio-temporal feature map. Based on the regular features of the real thermal radiation texture, the random interference of the pseudo-texture is suppressed;

[0113] Step S13, enhancing the real texture features and marking the confidence level: processing the texture data in the spatio-temporal feature map, and using the attention mechanism to strengthen the brightness and spectral characteristics of the real thermal radiation texture.

[0114] In the embodiments of the present invention, it needs to be further explained that the stress modeling module includes a multi-scale thermal stress coupling unit, which separates the microscopic and macroscopic thermal effect features based on the enhanced image sequence by using the multi-resolution decomposition technology; combines the microscopic defect distribution data of the copper pipe material to construct a multi-scale copper pipe thermal stress distribution model; uses the data fusion algorithm to integrate the microscopic stress field and the macroscopic stress field to generate a coupled thermal stress distribution map, improving the accuracy of the boundary of the cutting guidance area.

[0115] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for automatic copper tube cutting based on vision guidance, characterized in that: The following steps are involved: Step 1: Thermal radiation texture acquisition and feature extraction: The original thermal radiation texture image of the copper tube cutting area is collected by infrared camera, and the environmental parameters are recorded synchronously. The thermal radiation texture feature data set is extracted and the texture consistency is ensured through cross-stage correction; the thermal radiation texture feature data set includes spatial frequency map, dynamic change vector field and corresponding environmental parameter table; Step 2: Signal separation and enhancement: Separate and enhance the thermal radiation texture signal based on the thermal radiation texture feature dataset, generate an enhanced image sequence and annotate the confidence score of each texture feature; The confidence score is obtained by obtaining the instantaneous feature vector v_t, recording the average vector of the feature in the historical time series as v_avg, and using the formula Calculate the feature temporal consistency index C within the time window T; The probability P of each texture feature in the discriminator is extracted, and combined with the feature temporal consistency index C, the confidence score S is calculated by the following formula: ； Among them, α is the adjustment coefficient, which controls the impact of feature temporal consistency on the confidence score; Step 3: Thermal stress modeling and correction: Based on the enhanced image sequence and material properties, a multi-scale coupled thermal stress distribution model of the copper tube is constructed. The finite element analysis method is used to estimate the temperature field and stress field during the cutting process. The thermal stress distribution map is generated and updated in real time. The dynamic threshold is calculated based on the morphological evolution to guide the cutting path, and the boundary coordinates and adjustment log are output. Step 4: Adaptive cutting execution: Use the Bezier curve algorithm to generate an adaptive cutting path and execute it, adjust the log to optimize the cutting parameters according to the dynamic threshold, and generate a cutting report containing path coordinates, thermal radiation texture features and optimization parameters.

2. The method for automatic copper tube cutting based on vision guidance according to claim 1 is characterized in that: The step 1 includes a thermal radiation texture cross-stage correction step: The time series of the thermal radiation texture feature data set is clustered by the dynamic time warping algorithm to analyze the texture morphology evolution trend at different cutting stages. Based on the clustering results, the feature alignment technology is used to adjust the dynamically changing vector field so that the texture features across stages remain consistent, and a corrected thermal radiation texture feature dataset is generated and passed to the signal separation and enhancement step.

3. The method for automatic copper tube cutting based on vision guidance according to claim 1 is characterized in that: The step one includes an adaptive frame rate adjustment step, which adjusts the sampling frame rate of the infrared camera in real time according to the dynamic change speed of the thermal radiation texture or the environmental parameters of the cutting process to balance the data quality and processing efficiency.

4. The method for automatic copper tube cutting based on vision guidance according to claim 1 is characterized in that: In the process of acquiring the thermal stress distribution map, the cutting process environmental parameters are used to calibrate the parameters of the copper tube thermal stress distribution model. The enhanced image sequence is used as the initial condition to input the calibrated copper tube thermal stress distribution model. The real-time updated thermal stress distribution map is output through iterative calculation to reflect the temperature gradient and stress concentration area on the surface of the copper tube.

5. The method for automatic copper tube cutting based on vision guidance according to claim 1 is characterized in that: The step three includes a multi-scale thermal stress coupling step: based on the enhanced image sequence, a multi-resolution decomposition technology is used to separate the microscopic and macroscopic thermal effect characteristics; a multi-scale copper tube thermal stress distribution model is constructed in combination with the microscopic defect distribution data of the copper tube material; a data fusion algorithm is used to integrate the microscopic stress field and the macroscopic stress field to generate a coupled thermal stress distribution map, thereby improving the accuracy of the cutting guide area boundary.

6. The method for automatic copper tube cutting based on vision guidance according to claim 1 is characterized in that: It also includes an airflow noise management step, which starts real-time monitoring during the cutting process. When the auxiliary airflow acts on the surface of the copper tube, the airflow noise management step is triggered, including: Step S11, monitoring auxiliary airflow parameters and predicting noise characteristics: recording the airflow parameters of the auxiliary airflow, including the pressure and flow rate parameters of the auxiliary airflow; establishing a mapping relationship between airflow noise and image features to obtain a pseudo-texture prediction model, which is used to output the brightness and spectral characteristics of the pseudo-texture; Step S12, extracting and analyzing the spatiotemporal characteristics of texture: obtaining an infrared image sequence and a pseudo-texture prediction result, using a 3D convolutional neural network to analyze the spatiotemporal characteristics of the infrared image, generating a spatiotemporal feature map, based on the regular characteristics of the real thermal radiation texture, suppressing the random interference of the pseudo-texture; Step S13, enhancing real texture features and marking confidence: processing texture data in the spatiotemporal feature map, and using an attention mechanism to enhance the brightness and spectral characteristics of the real thermal radiation texture.

7. A method for automatic copper tube cutting based on vision guidance according to any one of claims 1 to 6, characterized in that: The dynamic threshold is obtained as follows: The enhanced image sequence is compared with the prediction results of the copper tube thermal stress distribution model to calculate the dynamic threshold of the thermal stress concentration area. The specific method is as follows: The boundary with the largest temperature gradient in the enhanced image is extracted, spatially matched with the stress concentration area predicted by the thermal stress distribution model of the copper tube, and the matching degree D is calculated; Based on the matching results, the dynamic threshold formula is defined as ,in is the reference temperature, γ is the gain coefficient, and H is the spatial complexity of the texture feature; γ is adjusted according to the texture feature confidence score S, and γ is positively correlated with the confidence score S; The time trend of the image sequence is enhanced by historical time series, and the Kalman filter is used to correct the instantaneous fluctuation, remove outliers, and generate a smoothed dynamic threshold sequence; the instantaneous random fluctuation of the thermal radiation texture is corrected by time series analysis, and outliers are removed.

8. The method for automatic copper tube cutting based on vision guidance according to claim 7 is characterized in that: The matching degree is calculated using the morphological evolution of the thermal stress concentration area, and the dynamic threshold is calculated as follows: ； ； in, is the second-order gradient of the image at time t, is the second-order gradient of the image at the initial moment, is the probability distribution of each frequency component in the spatial frequency diagram, and T is the time window.

9. A visually guided copper tube automatic cutting system, characterized in that: include: The thermal pattern acquisition module uses a high-frame-rate infrared camera to collect the original thermal radiation texture image of the copper tube cutting area in real time, and simultaneously records the environmental parameters of the cutting process; The feature extraction module uses wavelet transform technology to analyze the spatial frequency distribution and mark the thermal stress concentration area and high temperature gradient edge; Combined with the optical flow algorithm, the dynamic change vector of the thermal radiation texture is calculated, the evolution trend of the thermal radiation texture over time is quantified, the instantaneous and random characteristics are captured, and a thermal radiation texture feature data set is generated, including a spatial frequency map, a dynamic change vector field, and an environmental parameter table; The signal enhancement module performs image segmentation on the original thermal radiation texture image based on the thermal radiation texture feature dataset and extracts the effective thermal radiation texture signal; enhances the image clarity and continuity through the generative adversarial network, generates an enhanced image sequence, extracts texture features and annotates the confidence score for each feature; The confidence score is obtained by obtaining the instantaneous feature vector v_t, recording the average vector of the feature in the historical time series as v_avg, and using the formula Calculate the feature temporal consistency index C within the time window T; The probability P of each texture feature in the discriminator is extracted, and combined with the feature temporal consistency index C, the confidence score S is calculated by the following formula: ； Among them, α is the adjustment coefficient, which controls the impact of feature temporal consistency on the confidence score; The stress modeling module receives the enhanced image sequence, combines the copper tube material properties, builds the copper tube thermal stress distribution model, generates and updates the thermal stress distribution map in real time; marks the cutting guide area boundary based on the thermal stress distribution map, and adjusts the dynamic threshold to ensure cutting accuracy; outputs the real-time updated thermal stress distribution map, cutting guide area boundary coordinates and dynamic threshold adjustment log; The path optimization module uses the Bezier curve algorithm to generate an adaptive cutting path based on the thermal stress distribution map, the boundary coordinates of the cutting guide area and the dynamic threshold adjustment log; optimizes the cutting parameters in combination with the dynamic threshold adjustment log, executes the cutting operation and generates a cutting report.

Images