A welding arc detection method based on a pulse camera

Abstract

The application discloses a welding arc detection method based on a pulse camera, which comprises the following steps: using a pulse camera to capture a dynamic scene under an arc welding working scene and outputting corresponding pulse data flow; based on the pulse data flow, sequentially performing density enhancement, multi-scale integration and fusion reconstruction to obtain a high-quality reconstructed arc image; then analyzing and processing the high-quality reconstructed arc image to obtain arc geometric parameters and temperature field parameters; and finally, realizing quantitative evaluation of welding arc stability and optimization of welding quality according to the arc parameters. The method provided by the application can capture arc parameters reflecting the dynamic characteristics of the arc in the arc welding work by applying a pulse camera with time statistical characteristics, thereby providing support for welding state stability monitoring and welding quality detection and helping to monitor defects and abnormalities that are difficult to capture by a traditional camera.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent welding, and in particular to a welding arc detection method based on a pulse camera. BACKGROUND

[0002] In the field of modern industrial manufacturing, welding as a key connection process, its quality is directly related to the safety and durability of the product. The traditional welding quality detection method has been gradually replaced by advanced technology based on visual monitoring due to the limitations of efficiency and accuracy. Among them, the pulse camera uses its ultra-high-speed imaging advantage to provide a high-efficiency means for the monitoring and control of welding quality.

[0003] The traditional welding arc detection method has problems such as poor real-time performance, low precision and dependence on manual experience, which makes it difficult to comprehensively and accurately evaluate the stability of the arc, and the traditional sensor is easily limited in strong light and high temperature environment. In the prior art, the pulse camera image reconstruction method, such as the SSIR framework based on deep pulse neural network, emphasizes low energy consumption and time sequence processing, but has the limitation of single path integration and cannot flexibly process variable density pulses; the HSSR framework improves the identification of high-speed objects by correlating the features of long and short time pulse streams, but focuses on short-time feature extraction; the hierarchical time sequence filtering framework uses the time sequence correlation on the motion trajectory to suppress noise and blur, and realizes clear reconstruction through short-time and long-time filtering, but depends on supervision or motion estimation complexity. The present application proposes a multi-scale strategy, which balances global and local receptive fields by parallel processing of different scale window lengths, dynamically adapts to variable density pulse signals, significantly suppresses noise and improves reconstruction accuracy; reduces the computational cost and complexity, realizes efficient data compression and real-time processing; integrates cross-scale correlation, realizes global balance while preserving local details, improves image quality and application robustness. The pulse camera technology, with its high-speed imaging and strong anti-interference ability, can capture the arc shape and temperature distribution in real time, provide more accurate arc stability analysis, overcome the limitations of traditional methods, and is expected to become an important means of welding quality monitoring, promoting the optimization and improvement of welding process.

[0004] The pulse camera technology provides a new perspective in welding arc detection, which can provide clear images even in strong arc light and complex welding environment, and can capture transient phenomena in the welding process through high-speed imaging, enabling operators to accurately monitor welding quality. The arc cross-sectional area geometry recorded by the pulse camera and the calculated temperature field distribution can be used to detect the stability of the arc, thereby controlling the quality of the weld. SUMMARY

[0005] In view of the problems in the prior art, the present application provides a welding arc detection method based on a pulse camera, which introduces a pulse camera technology, utilizes the super-high-speed imaging feature of the pulse camera, captures transient phenomena in a welding process, and accurately monitors welding quality.

[0006] To achieve the above technical purposes, the present application provides the following technical solutions:

[0007] A welding arc detection method based on a pulse camera, which specifically comprises the following steps:

[0008] S1. In an electric arc welding scene, a pulse camera is introduced to capture a pulse data stream of high-speed dynamic changes of an electric arc in a welding process.

[0009] S2. The pulse data stream obtained in step S1 is subjected to integral pre-processing to obtain a density-enhanced pulse signal.

[0010] S3. The pulse data stream subjected to integral pre-processing is subjected to multi-scale path integral to obtain a multi-path integral result.

[0011] S4. The multi-path integral result is fused and reconstructed to obtain a high-quality reconstructed electric arc image.

[0012] S5. The high-quality electric arc reconstructed image is subjected to threshold segmentation and image filling operations.

[0013] S6. The image after the segmentation operation is further subjected to electric arc edge detection and geometric parameter extraction; the geometric parameters include an electric arc area and a perimeter.

[0014] S7. The electric arc image after the edge detection is subjected to temperature field parameter extraction.

[0015] S8. The electric arc stability is analyzed by combining the obtained electric arc geometric parameters and temperature field parameters.

[0016] Further, step S2 specifically comprises:

[0017] S21. Input an original binary pulse data stream , wherein is a pixel coordinate, is a time index; a preliminary intensity estimate is calculated, and the formula is represented as:

[0018] ;

[0019] wherein, is a trigger threshold, is a pulse interval;

[0020] S22. The preliminary intensity estimate is processed using a 3*3 convolution kernel to output a density map , and the formula is represented as:

[0021] ;

[0022] wherein, is a neighborhood; m represents each element in the neighborhood is a learnable weight;

[0023] S23, applying a ReLU activation function to the density map to ensure all values are non-negative, obtaining a density-enhanced impulse signal , which is expressed by the formula:

[0024] ;

[0025] The size of is the same as the input, which is used for subsequent multi-scale integration.

[0026] Further, step S3 specifically comprises:

[0027] S31, inputting the density-enhanced impulse signal , adopting multiple independent paths, each path corresponding to a different integration scale;

[0028] S32, for each path, calculating an adaptive kernel , which is expressed by the formula:

[0029] ;

[0030] ;

[0031] wherein, is a global density, , is a path-specific parameter, which is learned through backpropagation, is a density map, is a pixel coordinate, is a time index;

[0032] The integration result of each path is:

[0033] ;

[0034] wherein, j is a path index; is a window size of the jth path; i is a traversal index, representing a relative time offset within the time sequence window, and the value ranges from 1 to ;

[0035] ​S33, add 1x1 convolution after each integral path to fuse spatial neighborhood and use activation function Softplus to ensure smooth accumulation, output the final multi-path smooth integral result The size of each integral result is HxWxT, H, W and T represent image height, image width and time step number respectively; different accumulation strategies are explored in parallel to ensure adaptation to high and low density pulses.

[0036] Further, step S4 specifically includes:

[0037] S41, input multi-path smooth integral result , calculate the intensity variance of each path , the formula is:

[0038] ;

[0039] Where, is the intensity mean value; is the pixel coordinate, is the time index;

[0040] S42, select the path with the lowest variance as the main path, and other paths are fused by weighting, the formula is:

[0041] ;

[0042] Where, is the weighted fusion image; is the weighting coefficient of the jth path, and its calculation formula is:

[0043] ;

[0044] Where, is an exponential decay function, is a learnable scaling parameter used to control the sensitivity of the weight;

[0045] S43, apply 1x1 convolution to fuse channel information, then use Sigmoid function normalization to ensure that the output intensity value is in the range of [0, 1], avoid overflow and enhance nonlinear mapping, generate high-quality reconstructed arc image with low noise and high consistency , the formula is:

[0046] ;

[0047] ;

[0048] Where, is the convolution weight, b is the bias, and c is the channel index, is the channel fusion information.

[0049] Further, step S5 specifically comprises:

[0050] S51, threshold segmentation; setting a gray threshold , segmenting the arc region from the background to obtain a binary image ;

[0051] S52, image filling; using a closing operation to fill the binary image to obtain a complete filled image .

[0052] Further, step S6 specifically comprises:

[0053] S61, applying Sobel operator for edge detection, finding edge position by calculating image gradient and gradient amplitude;

[0054] S62, after accurately identifying the edge, extracting the geometric parameters of the arc by analyzing the edge information and applying geometric calculation method; the geometric parameters include: arc column longitudinal section, circumference and area, wherein the area is calculated by the number of foreground pixels in the binary image; the circumference is calculated by the number of edge pixels obtained by edge detection.

[0055] Further, step S7 specifically comprises:

[0056] S71, calculating the projection radiation intensity, assuming that the welding arc is axisymmetric distribution, extracting the distribution of side projection intensity from the arc profile obtained by edge detection , wherein r is the radial distance from the arc axis to the edge; the projection intensity is the line integral radiation intensity obtained by averaging the image intensity along the arc axis direction, which is expressed as:

[0057] ;

[0058] wherein, is the length of the arc axis, refers to the time index , the image intensity value along the axis coordinate , the radial coordinate r perpendicular to the axis;

[0059] S72, applying Abel inverse transform to calculate radial emissivity, using Abel inverse transform to recover radial emissivity from the projection intensity , which represents the volume emissivity of the arc at the radial position r, and is expressed as:

[0060] ;

[0061] wherein, is the integral variable; the upper limit of integration ∞ is replaced by the arc radius R in practical calculation, and a numerical integration method is used for discrete calculation to obtain the radial emissivity of each discrete point :

[0062] ;

[0063] wherein, p represents the discrete point index; is the radial step size, and N is the number of discrete points;

[0064] S73, calculating the radial temperature distribution: based on the Stefan-Boltzmann law, the radial emissivity is converted into the radial temperature ; assuming that the arc is a blackbody or a gray body, the calculation formula of is:

[0065] ;

[0066] wherein, is the Stefan-Boltzmann constant, and the temperature unit is Kelvin.

[0067] Further, step S8 specifically comprises:

[0068] S81, calculating the geometric feature time series change rate by using the area and the perimeter of the continuous frame image, to quantify the stability of the arc shape; the geometric feature time series change rate includes the area change rate and the perimeter change rate , which is defined as the relative difference between the current frame and the previous frame, and the formula is expressed as:

[0069] ;

[0070] wherein, is the arc area of the t-th frame, with the unit of pixel²; is the arc area of the previous frame; is the arc perimeter of the t-th frame, with the unit of pixel; is the arc perimeter of the previous frame; |·| represents the absolute value;

[0071] S82, calculating the temperature field uniformity; based on the radial temperature , the temperature standard deviation and the temperature gradient are calculated to quantify the stability of the heat distribution; the formula is expressed as:

[0072] ;

[0073] ;

[0074] wherein N is the number of radial discrete points; is the radial temperature at the p th discrete point; is the average radial temperature, ; is the radial step length;

[0075] S83, quantifying the stability index; integrating the geometric and temperature features to calculate the overall stability index , evaluating the arc stability; the formula is expressed as:

[0076] ;

[0077] wherein, is the multi-frame average area change rate; is the multi-frame average perimeter change rate; 、 、 is the weight coefficient; the arc stability threshold is set to determine whether the welding parameters need to be adjusted.

[0078] Based on the above technical solutions, the present application has the following beneficial effects:

[0079] The method proposed by the present application introduces the pulse camera technology, combines the ultra-high-speed imaging features, captures the pulse data of the welding arc in the welding arc working process, and obtains a clear image after multi-scale integral reconstruction processing. Then, through a series of processes such as edge detection, image optimization, feature extraction and temperature field calculation, the geometric parameters and temperature field distribution of the arc are finally output. These results can intuitively reflect the dynamic characteristics of the arc in the welding process, provide important data support for welding quality control, process optimization and equipment design, and can monitor defects and abnormalities that are difficult to capture by traditional cameras. BRIEF DESCRIPTION OF DRAWINGS

[0080] Fig. 1 is the overall flowchart of the welding arc detection method based on the pulse camera proposed by the present application;

[0081] Fig. 2 is the specific process schematic diagram of the method proposed by the present application. DETAILED DESCRIPTION

[0082] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.

[0083] The steps in the present application are arranged with labels, but are not used to limit the order of the steps, unless the order of the steps is explicitly stated or the execution of a certain step needs other steps as a basis, otherwise the relative order of the steps can be adjusted. It can be understood that the term "and / or" used herein relates to and covers any and all possible combinations of one or more of the associated listed items.

[0084] As Figs. 1-2 shown, the present application proposes a welding arc detection method based on pulse camera, which specifically includes the following steps:

[0085] S1, in the arc welding scene, introduce pulse camera, capture the high-speed dynamic change of pulse data stream in the welding process; these high-speed changing pulse images are used as the basis for in-depth analysis of the stability of the welding process.

[0086] S2, the integral pre-processing of the pulse data stream obtained in step S1 is carried out, and the density enhanced pulse signal is obtained;

[0087] As a preferred embodiment, step S2 specifically includes:

[0088] S21, input the original binary pulse data stream , wherein is the pixel coordinate, is the time index; calculate the preliminary intensity estimate, the formula is represented as:

[0089] ;

[0090] wherein, is the trigger threshold, is the pulse interval; this step generates the initial image with noise;

[0091] S22, using 3*3 convolution kernel to process the preliminary intensity estimate, output the density map , the formula is represented as:

[0092] ;

[0093] wherein, is the neighborhood; m represents each element in the neighborhood is the learnable weight;

[0094] S23, apply ReLU activation function to the density map , ensure all values are non-negative, get the density enhanced pulse signal , the formula is represented as:

[0095] ; ​

[0096] The size of the input is the same as the input, which is used for subsequent multi-scale integration.

[0097] The present embodiment suppresses the initial quantization noise generated in step S21 by density smoothing, thereby reducing the variance of the overall image data.

[0098] S3, multi-scale path integration is performed on the pulse data stream after the integration preprocessing, and a multi-path integration result is obtained;

[0099] As a preferred embodiment, step S3 specifically includes:

[0100] S31, input the density-enhanced pulse signal , adopt multiple independent paths, each path corresponding to different integration scales (calculate the instantaneous light intensity by accumulating the signals of several consecutive time steps in the time dimension through different time window lengths, and each time step corresponds to a sampling period of the camera); In the design of the present application, three independent paths are adopted, which are 50-step integration scale of short time window, 20-step integration scale of medium time window and 50-step integration scale of long time window;

[0101] In the present embodiment, multiple paths are designed for integration, which on one hand improves robustness and adaptability: by parallel processing of different scale window lengths, balancing global and local receptive fields, dynamically adapting to variable density pulse signals (such as welding arc strong light interference), significantly suppressing noise and improving reconstruction accuracy; On the other hand, it reduces the computational cost and complexity: multi-scale parallel fusion expands the receptive field while maintaining linear complexity, avoids single-scale limitations, realizes efficient data compression and real-time processing; At the same time, it also enhances detail preservation and balance: integrates cross-scale correlation (such as spatiotemporal aggregation), preserves local details while achieving global balance (such as temperature field uniformity), improves image quality and application robustness;

[0102] S32, for each path, calculate the adaptive kernel , the formula is:

[0103] ;

[0104] ;

[0105] wherein, is the global density, , is a path-specific parameter, learned by backpropagation, is a density map, is a pixel coordinate, is a time index;

[0106] The integration result of each path is:

[0107] ;

[0108] where j is the path index; is the window size of the jth path; i is the traversal index, representing the relative time offset within the timing window, ranging from 1 to ;

[0109] S33, a 1x1 convolution is added after each integral path to fuse the spatial neighborhood and use the activation function Softplus to ensure smooth accumulation, outputting the final multi-path smooth integral result , the size of each integral result is HxWxT (H, W, and T represent image height, image width, and time step number, respectively); different accumulation strategies are explored in parallel to ensure adaptation to high and low density pulses.

[0110] S4, the multi-path integral results are fused and reconstructed to obtain a high-quality reconstructed arc image;

[0111] As a preferred embodiment, step S4 specifically includes:

[0112] S41, input the multi-path smooth integral results , calculate the intensity variance of each path , the formula is:

[0113] ;

[0114] where, is the intensity mean; is the pixel coordinate, is the time index;

[0115] S42, select the path with the lowest variance as the main path, and other paths are fused with weights, the formula is:

[0116] ;

[0117] where, is the weighted fusion image; is the weighting coefficient of the jth path, and its calculation formula is:

[0118] ;

[0119] where, is an exponential decay function, is a learnable scaling parameter, used to control the sensitivity of the weight.

[0120] In this embodiment, the best integration result is selected and fused in the multi-path integration result to realize soft decision fusion, ensure the advantages of the final output comprehensive multi-path, and prefer low-noise paths to improve the timing consistency and overall quality of the reconstruction.

[0121] S43, apply 1x1 convolution to fuse channel information, then use Sigmoid function normalization to ensure that the output intensity value is in the range of [0, 1], avoid overflow and enhance nonlinear mapping, and generate high-quality reconstructed arc images with low noise and high consistency , which is expressed by the formula:

[0122] ;

[0123] ;

[0124] wherein, is the convolution weight, b is the bias, and c is the channel index, is the channel fusion information.

[0125] This ensures that the output intensity value of is in the range of [0, 1], avoids overflow, and enhances nonlinear mapping to generate a final image sequence with low noise and high consistency.

[0126] S5, threshold segmentation and image filling operations are performed on the high-quality arc reconstruction image.

[0127] As a preferred embodiment, step S5 specifically includes:

[0128] S51, threshold segmentation; set a gray threshold to segment the arc region from the background and obtain a binary image , which is expressed by the formula:

[0129] ;

[0130] wherein, is the gray value corresponding to the coordinate in the image;

[0131] S52, image filling; use the closing operation to fill the binary image to obtain a complete filled image , which is expressed by the formula:

[0132] ;

[0133] wherein, is a structural element; is the dilation operation, is the erosion operation.

[0134] The step S5 directly acts on the high-quality image reconstructed in step S4 in this embodiment to generate a binary and complete arc region image, which is used as subsequent input; at the same time, threshold segmentation and image filling are performed to remove noise and interference, simplify data structure, and improve precision and robustness.

[0135] S6, further arc edge detection and geometric parameter extraction are performed on the image after the segmentation operation; the geometric parameters include arc area and perimeter;

[0136] As a preferred embodiment, the step S6 specifically includes:

[0137] S61, a Sobel operator is applied for edge detection, and edge positions are found by calculating gradients and gradient amplitudes of the image; a calculation formula of the Sobel edge detection is as follows:

[0138] ;

[0139] ;

[0140] wherein, and respectively represent gradients of the image in and directions;

[0141] the gradient amplitude of the edge intensity is calculated according to the following formula:

[0142] ;

[0143] In this embodiment, by calculating the gradients and gradient amplitudes of the image, the Sobel operator can help find regions with significant changes in grayscale values in the image, that is, possible edges of objects. Since the welding arc itself usually has a large grayscale difference (higher or lower brightness) from the background, the Sobel operator can effectively identify the edges of the arc by reflecting such significant changes.

[0144] S62, after the edges are accurately identified, geometric parameters of the arc are extracted by analyzing the edge information and applying a geometric calculation method; the geometric parameters include: arc column longitudinal section, perimeter and area, wherein the area is calculated by the number of foreground pixels in the binary image; the perimeter is calculated by the number of edge pixels obtained by edge detection; and a specific calculation process is as follows:

[0145] Area A is calculated by the number of foreground pixels in the binary image, and a formula is as follows:

[0146] ;

[0147] wherein, represents the pixel value in the binary image, if 1, it means that the pixel belongs to the arc region, if 0, it does not belong to;

[0148] Circumference C The number of edge pixels obtained by edge detection is calculated, and the formula is:

[0149] ;

[0150] Wherein, represents the pixel representing the arc edge in the edge detection result.

[0151] S7, temperature field parameter extraction is performed on the arc image after edge detection;

[0152] As a preferred embodiment, step S7 specifically includes:

[0153] S71, calculate the projection radiation intensity, assuming that the welding arc is axisymmetric distribution, from the arc profile obtained by edge detection, the side projection intensity Distribution is extracted, where r is the radial distance from the arc axis to the edge; The projection intensity Is the line integral radiation intensity obtained by averaging the image intensity along the arc axis direction, and the formula is:

[0154] ;

[0155] Wherein, is the length of the arc axis, Refers to the time index Below, along the axis coordinate , the image intensity value of the radial coordinate r perpendicular to the axis;

[0156] S72, apply Abel inverse transform to calculate radial emissivity, use Abel inverse transform to recover radial emissivity From the projection intensity , which represents the volume emissivity of the arc at the radial position r, and the formula is:

[0157] ;

[0158] Wherein, Is the integral variable; The upper limit of integration ∞ is replaced by the arc radius R in actual calculation, and numerical integration method is used for discrete calculation, to obtain the radial emissivity Of each discrete point:

[0159] ;

[0160] Wherein, , p represents the index of discrete points, which takes values from 0 to N-1 here; is the radial step length, N is the number of discrete points;

[0161] S73, calculate the radial temperature distribution: based on the Stefan-Boltzmann law, the radial emissivity is converted to the radial temperature ; assuming that the arc is a blackbody or gray body, the calculation formula is:

[0162] ;

[0163] wherein, is the Stefan-Boltzmann constant, the temperature unit is in Kelvin.

[0164] S8, combine the obtained arc geometric parameters and temperature field parameters to perform arc stability analysis.

[0165] As a preferred embodiment, step S8 specifically includes:

[0166] S81, calculate the geometric feature time series change rate using the area and perimeter of the continuous frame image to quantify the stability of the arc shape; the geometric feature time series change rate includes the area change rate and the perimeter change rate , defined as the relative difference between the current frame and the previous frame, the formula is expressed as:

[0167] ;

[0168] wherein, is the arc area of the t-th frame, in units of pixels²; is the arc area of the previous frame; is the arc perimeter of the t-th frame, in units of pixels; is the arc perimeter of the previous frame; |·| represents the absolute value; this step evaluates the shape fluctuation, and the smaller the change rate indicates that the arc geometry is more stable, which is suitable for detecting flickering or deviation.

[0169] S82, calculate the temperature field uniformity; based on the radial temperature , calculate the temperature standard deviation and the temperature gradient to quantify the stability of the heat distribution; the formula is expressed as:

[0170] ;

[0171] ;

[0172] wherein, N is the number of radial discrete points; is the radial temperature at the p th discrete point;​ is the average radial temperature, ; is the radial step; this step evaluates thermal non-uniformity, is smaller or is more stable, avoiding overheating or cold spots leading to defects.

[0173] S83, quantifying stability index; combining geometric and temperature features, calculating overall stability index , evaluating arc stability; formula is:

[0174] ;

[0175] wherein, is the multi-frame average area change rate; is the multi-frame average perimeter change rate; , , is the weight coefficient; set the arc stability threshold to determine whether to adjust the welding parameters. range [0, 1], close to 1 indicates high stability. In this embodiment, the threshold is set to determine: if , it is marked as unstable, triggering welding parameter adjustment.

[0176] In summary, the welding arc detection method based on pulse camera proposed in the present application introduces a pulse camera with time statistical characteristics to capture arc parameters reflecting the dynamic characteristics of the arc in the arc welding process, providing support for state stability monitoring and welding quality detection, and helping to monitor defects and abnormalities that are difficult to capture by traditional cameras.

[0177] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or features of different embodiments or examples described in the present specification without contradiction.

[0178] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or a combination of the above.

[0179] The above embodiments introduce the present application in detail, the principles and embodiments of the present application are described by applying specific examples, the above example is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific embodiment and application range will be changed, and the above description should not be understood as the limitation of the present application.

Claims

1. A method for detecting welding arcs based on a pulse camera, characterized in that, Includes the following steps: S1. In the arc welding scenario, a pulse camera is introduced to capture the pulse data stream of the arc changing rapidly during the welding process; S2. Perform pre-integration processing on the pulse data stream obtained in step S1 to obtain a density-enhanced pulse signal; S3. Perform multi-scale path integration on the pulse data stream after pre-integration processing to obtain the multi-path integration result; S4. The multipath integration results are fused and reconstructed to obtain a high-quality reconstructed arc image; S5. First, perform threshold segmentation and image filling operations on the high-quality arc reconstruction image; S6. Further perform arc edge detection and geometric parameter extraction on the segmented image; the geometric parameters include the arc area and perimeter. S7. Extract temperature field parameters from the arc image after edge detection; S8. Using the combined arc geometric parameters and temperature field parameters, perform arc stability analysis, specifically including: S81. Using the area and perimeter of consecutive frame images, calculate the temporal change rate of geometric features to quantify the stability of the arc shape; the temporal change rate of geometric features includes the area change rate. and the rate of change of perimeter , defined as the relative difference between the current frame and the previous frame, is expressed by the formula: ； in, It is the arc area in frame t, in pixels². It is the arc area of ​​the previous frame; It is the arc circumference in frame t, in pixels; It represents the arc circumference of the previous frame; |·| represents the absolute value; S82. Calculate the temperature field uniformity; based on radial temperature. Calculate the standard deviation of temperature and temperature gradient This is used to quantify the stability of heat distribution; the formula is expressed as: ； ； Where N is the number of radial discrete points; It is the first p Radial temperature at discrete points; It is the average radial temperature. ; Radial step size; S83, Quantitative Stability Index; Calculate the overall stability index by combining geometric and temperature characteristics. To assess arc stability; the formula is expressed as: ； in, It is the average area change rate across multiple frames; It is the average perimeter change rate across multiple frames; , , It is a weighting coefficient; it sets the arc stability threshold to determine whether welding parameters need to be adjusted.

2. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S2 specifically includes: S21. Input raw binary pulse data stream ,in For pixel coordinates, For time indexing; calculate the preliminary intensity estimate, expressed by the formula: ； in, It is the trigger threshold. It is the pulse interval; S22. Use a 3×3 convolution kernel to process the initial intensity estimation and output the density map. The formula is expressed as: ； in, It represents the neighborhood; m represents the neighborhood. Each element in; These are learnable weights; S23, Density Map By applying the ReLU activation function and ensuring that all values ​​are non-negative, a density-enhanced pulse signal is obtained. The formula is expressed as: ； The size is the same as the input, and it is used for subsequent multi-scale integration.

3. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S3 specifically includes: S31, Input density enhancement pulse signal Multiple independent paths are used, each corresponding to a different integration scale; S32. For each path, calculate the adaptive kernel. The formula is expressed as: ； ； in, It is global density. , Path-specific parameters are learned through backpropagation. For density plots, For pixel coordinates, For time indexing; The integral result for each path for: ； Where j is the path index; `i` is the window size for the `j`-th path; `i` is the traversal index, representing the relative time offset within the time window, ranging from 1 to `i`. ; S33. After each integration path, add a 1×1 convolution to fuse the spatial neighborhood, and use the Softplus activation function to ensure smooth accumulation, outputting the final multi-path smooth integration result. Each integration result has a size of H×W×T, where H, W, and T represent the image height, image width, and time step, respectively; different accumulation strategies are explored in parallel to ensure adaptation to high and low density pulses.

4. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S4 specifically includes: S41. Input the multipath smoothing integral result. Calculate the strength variance for each path. The formula is expressed as: ； in, The average intensity; For pixel coordinates, For time indexing; S42. Select the path with the lowest variance as the main path, and then weight and merge the other paths. The formula is as follows: ； in, For weighted fused images; It is the weighting coefficient for the j-th path, and its calculation formula is: ； in, It is an exponentially decaying function. It is a learnable scaling parameter used to control the sensitivity of the weights; S43. Apply 1×1 convolution to fuse channel information, then use the Sigmoid function for normalization to ensure that the output intensity value is within the range of [0,1], generating a high-quality reconstructed arc image with low noise and high consistency. The formula is expressed as: ； ； in, b is the convolution weight, c is the bias, and 'c' is the channel index. Information is integrated into the channel.

5. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S5 specifically includes: S51, Threshold segmentation; Set grayscale threshold. The arc region is segmented from the background to obtain a binarized image. ; S52, Image Filling: The closing operation is used to fill the binarized image to obtain a complete filled image. .

6. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S6 specifically includes: S61. Apply the Sobel operator for edge detection, and find the edge location by calculating the gradient and gradient magnitude of the image; S62. After accurately identifying the edges, the geometric parameters of the arc are extracted by analyzing the edge information and applying geometric calculation methods. The geometric parameters include: the longitudinal section of the arc column, the perimeter, and the area, where the area is calculated by the number of foreground pixels in the binarized image; the perimeter is calculated by the number of edge pixels obtained by edge detection.

7. The welding arc detection method based on a pulse camera according to claim 1, characterized in that, Step S7 specifically includes: S71. Calculate the projected radiation intensity. Assuming the welding arc is axisymmetrically distributed, extract the side projection intensity from the arc profile obtained by edge detection. The distribution of , where r is the radial distance from the arc axis to the edge; projected intensity The line integral radiation intensity, obtained by averaging the image intensity along the arc axis, is expressed by the formula: ； in, It is the length of the electric arc axis. This refers to the time index. Below, along the axis coordinates The image intensity value at the radial coordinate r perpendicular to the axis; S72. Calculate radial emissivity using the inverse Abel transform, and then use the inverse Abel transform to calculate the radial emissivity from the projected intensity. Restore radial emissivity It represents the volume emissivity of the electric arc at a radial position r, expressed by the formula: ； in, It is the integral variable; the upper limit of integration ∞ is replaced by the arc radius R in actual calculations, and the radial emissivity at each discrete point is obtained by discretization using numerical integration methods. : ； in, p represents the discrete point index; is the radial step size, and N is the number of discrete points; S73. Calculate the radial temperature distribution: Based on the Stefan-Boltzmann law, the radial emissivity... Convert to radial temperature Assuming the electric arc is a blackbody or a graybody, then The calculation formula is: ； in, It is the Stefan-Boltzmann constant, and the temperature unit is Kelvin.

Images