Offset detection method for new energy battery pole welding

Abstract

The invention provides an offset detection method for new energy battery pole welding, which comprises the following steps: inputting pole features and to-be-welded contour features contained in a reference image into a reference coordinate system for feature positioning to generate a reference feature library, collecting a dynamic image of a welding area in real time, and performing time sequence trend analysis; the reference feature library is used for conducting feature proofreading on the prediction result, when the proofreading result is abnormal, the pole abnormal state and the welding contour abnormal state of the current welding plate are recognized in the dynamic image, the current welding abnormal position corresponding to the current welding plate is determined, local parameter adjustment is conducted on the welding process of the current welding plate, and the welding quality of the current welding plate is improved. The abnormal rule of the welding area is excavated, the abnormal rule is used for overall parameter adjustment of all the welding processes, the defect that common problems cannot be radically solved through traditional local correction is overcome, the stability and reliability of overall welding production are greatly improved, and a core technical guarantee is provided for large-scale and standardized production.

Description

Technical Field

[0001] This invention relates to the field of new energy battery manufacturing technology, and in particular to a method for detecting offset in the welding of new energy battery terminals. Background Technology

[0002] In the manufacturing process of new energy batteries, electrode welding is one of the key processes, and the welding quality directly affects the battery's electrical performance, safety performance, and service life. During electrode welding, factors such as the precision of welding equipment, fluctuations in the welding environment, and positioning deviations of the board to be welded can easily lead to welding defects such as electrode misalignment and deviations of the welding contour from the preset trajectory. If these defects are not detected and corrected in time, they can result in poor contact of the battery electrodes, increased internal resistance, and even safety hazards such as short circuits and thermal runaway.

[0003] Existing technologies for detecting welding misalignment of new energy battery terminals have the following main shortcomings: First, traditional testing methods mostly adopt a post-event sampling inspection mode, which cannot monitor the offset during the welding process in real time. This results in defective products being unable to be traced and remedied in a timely manner after they leave the factory, increasing production costs. Second, the detection process is easily affected by the metallic luster reflection of the pole, resulting in inaccurate feature extraction and low offset detection accuracy. Third, the system can only make simple parameter adjustments to address the offset defect, lacking pattern mining based on full-process data and overall process optimization, thus failing to reduce the occurrence rate of offset defects from the root cause. Fourth, when an anomaly is detected, it is difficult to quickly locate the cause of the anomaly, and normal welding processes are not effectively reused, so production efficiency needs to be improved.

[0004] Therefore, the present invention provides a method for offset detection in welding of new energy battery terminals. Summary of the Invention

[0005] This invention provides a method for offset detection in welding of new energy battery terminals, enabling real-time and accurate detection of offset during the welding process, identifying the cause of the anomaly and correcting it in a timely manner. At the same time, it optimizes the overall welding process by mining anomaly patterns based on full-process data, thereby improving welding quality and production efficiency and reducing production costs.

[0006] This invention provides a method for detecting offset in welding terminals of new energy batteries, comprising: Step 1: Construct a reference coordinate system based on the reference image of the board to be welded, input the polar features and the contour features to be welded contained in the reference image into the reference coordinate system for feature localization, and generate the reference feature library of the board to be welded. Step 2: Real-time acquisition of dynamic images of the welding area, time-series trend analysis of the dynamic images to obtain the prediction results of the current welding plate, and feature verification of the prediction results using the benchmark feature library to obtain the verification results; Step 3: When the calibration result is abnormal, identify the abnormal state of the pole post and the abnormal state of the welding contour of the current welding plate in the dynamic image, and determine the current welding abnormality position corresponding to the current welding plate. Step 4: Generate a welding control signal based on the welding offset corresponding to each current welding abnormality position, and perform local parameter adjustment on the welding process of the current welding plate; Step 5: Record on-site data throughout the welding process to build a full-process database. In the full-process database, discover abnormal patterns in the welding area and use these abnormal patterns to adjust the parameters of all welding processes as a whole.

[0007] In one feasible embodiment, step 1 includes: Step 11: Extract the metallic luster features contained in the reference image, input the reference image into the polarization filter mapping space, iteratively adjust the polarization direction and polarization degree of the polarization filter mapping space until the feature value of the metallic luster features is lower than a specified threshold, and then optimize the grayscale value of the mapped reference image to obtain a high-quality reference image. Step 12: Select several stable feature points contained in the high-quality reference image, randomly select a number of stable feature points to construct a feature point set, use the least squares method to fit and calculate each feature point set, determine the X-axis direction and Y-axis direction of the reference coordinate system based on the calculation results, and establish the reference coordinate system with the geometric center of the plate to be welded as the origin. Step 13: Perform edge detection on the high-quality reference image to obtain several polar features contained in the reference image, and use the active contour model algorithm to perform contour tracking on the region to be welded in the high-quality reference image to obtain the contour features to be welded. Step 14: Input each of the pole features and each of the weldable contour features into the reference coordinate system to obtain the pole positioning information and pole information entropy of the weldable plate, as well as the weld contour positioning information and weld information entropy. Derive several execution standards of the weldable plate and generate the corresponding reference feature library.

[0008] In one feasible embodiment, step 2 includes: Step 21: Obtain the welding operation progress signal of the welding area. When the progress signal triggers the start of welding, construct a frequency dynamic acquisition scheme based on the welding head movement speed signal, acquire real-time welding images of the welding head, and statistically analyze all the real-time welding images to construct a dynamic image of the welding area. Step 22: Construct a time-series image sequence from the continuously acquired dynamic images in chronological order, extract the welding area feature information contained in each dynamic image using a convolutional neural network, capture the time dependency relationship between different welding area feature information to construct the trend curve of the welding area, and determine the welding stability of the current welding process. Step 23: Based on the welding stability, determine the offset prediction range of the current welding plate to obtain several offset results, filter the prediction results with the highest matching degree with the trend curve, and identify several prediction pole features and prediction welding profile features contained in the prediction results. Step 24: Use a weighted summation method to fuse the predicted pole feature with each pole feature in the benchmark feature library, and fuse the predicted welding contour feature with each welding contour feature in the benchmark feature library to obtain the comprehensive similarity and comprehensive overlap, and generate the verification result.

[0009] In one feasible embodiment, step 3 includes: Step 31: When the calibration result is abnormal, identify the preliminary information of the abnormal features contained in the calibration result, and at the same time separate the welding area and non-welding area in the dynamic image. In the separated welding area image, slide to match the abnormal feature template and mark the area with a matching degree higher than the specified matching value as the suspected abnormal area. Step 32: Construct a lightweight convolutional neural network model. Input the local image of the suspected abnormal region into the extraction layer, feature enhancement layer and classification recognition layer of the lightweight convolutional neural network model for feature extraction. Determine the pole abnormality state of the current welding plate based on the output parameters of each layer. Step 33: Use the subpixel edge detection algorithm to extract edges from the local image to obtain the subpixel level precise contour of the welding contour. Then, accurately match the subpixel level precise contour with the reference features of the contour to be welded in the reference feature library to determine the abnormal state of the welding contour of the current welding plate. Step 34: Locate the first image position corresponding to each pole state and the second image position corresponding to each welding contour state in the dynamic image, input the dynamic image into the reference coordinate system to identify each image position, and determine the current welding abnormality position of the current welding plate.

[0010] One feasible approach also includes: Identify the pre-welding information and post-welding information corresponding to each current welding anomaly location; The post-welding information is analyzed using the reference feature library to determine and display the cause of the welding abnormality corresponding to the current welding abnormality location.

[0011] In one feasible embodiment, step 4 includes: Step 41: Extract the pole column abnormality state parameters and welding profile abnormality state parameters corresponding to each current welding abnormality position, and construct the pole column offset vector and profile offset vector of the current welding abnormality position by combining the pole column offset direction and welding profile offset direction corresponding to the current welding position. Step 42: Determine the parameter adjustment direction and parameter adjustment range of the current welding abnormality position based on the pole offset vector and the contour offset vector, and perform PID adjustment on the parameter adjustment direction and parameter adjustment range corresponding to the same current welding plate to obtain the optimal parameter adjustment scheme corresponding to each current welding abnormality position; Step 43: Generate a standardized digital signal corresponding to the current welding anomaly position according to the optimal parameter adjustment scheme, perform anti-interference processing on the standardized digital signal to obtain the welding control signal corresponding to each current welding position, and transmit it to the welding area; Step 44: Locate the welding equipment corresponding to the current welding plate in the welding area, and use the welding control signal to locally adjust the welding process of the welding equipment.

[0012] In one feasible embodiment, step 5 includes: Step 51: Collect the offset data, control parameters and imaging information of the entire welding process and perform standardization processing respectively to build a full process database. Locate the full data corresponding to each local parameter adjustment process in the full process database, and set corresponding reason labels for the full data based on the parameter adjustment reason corresponding to each local parameter adjustment. Step 52: Mine several related data corresponding to each full data in the full process database, and logically sort out the related data corresponding to the full data with the same cause label to obtain the abnormal cause corresponding to each parameter tuning cause; Step 53: Simulate each of the above-mentioned abnormal causes under different welding scenarios, determine the abnormal patterns corresponding to each of the above-mentioned abnormal causes, analyze the abnormal manifestations of the abnormal patterns on each of the above-mentioned welding processes, and optimize each of the above-mentioned abnormal manifestations. Step 54: Based on the optimization results, confirm the content to be optimized for each welding process, and perform overall parameter adjustment for each welding process.

[0013] One feasible approach also includes: When the calibration result is normal, obtain the current process being executed for the current welding plate; Add a recommendation weight to the currently executed process; The recommended process is sequentially recommended to each welding area based on the recommended weight.

[0014] One feasible approach also includes: When the magnitude of the first vector corresponding to the pole offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. When the second vector magnitude corresponding to the contour offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. The defective products are then transported to the scrap area to await destruction.

[0015] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram illustrating the workflow of an offset detection method for welding terminals of new energy batteries in an embodiment of the present invention. Figure 2 This is a schematic diagram of the workflow of step 1 of an offset detection method for welding terminals of new energy batteries in an embodiment of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] Example 1: This example provides a method for detecting offset in the welding of electrode posts for new energy batteries, such as... Figure 1 As shown, it includes: Step 1: Construct a reference coordinate system based on the reference image of the board to be welded, input the polar features and the contour features to be welded contained in the reference image into the reference coordinate system for feature localization, and generate the reference feature library of the board to be welded. Step 2: Real-time acquisition of dynamic images of the welding area, time-series trend analysis of the dynamic images to obtain the prediction results of the current welding plate, and feature verification of the prediction results using the benchmark feature library to obtain the verification results; Step 3: When the calibration result is abnormal, identify the abnormal state of the pole post and the abnormal state of the welding contour of the current welding plate in the dynamic image, and determine the current welding abnormality position corresponding to the current welding plate. Step 4: Generate a welding control signal based on the welding offset corresponding to each current welding abnormality position, and perform local parameter adjustment on the welding process of the current welding plate; Step 5: Record on-site data throughout the welding process to build a full-process database. In the full-process database, discover abnormal patterns in the welding area and use these abnormal patterns to adjust the parameters of all welding processes as a whole.

[0020] In this example, the reference image is the initial image of the board to be welded used to construct the detection reference. It contains standard features of the poles and the contour to be welded and serves as the reference basis for subsequent feature calibration. In this example, the reference coordinate system is a two-dimensional coordinate system constructed based on the reference image. It takes the geometric center of the plate to be welded as the origin and determines the X and Y axis directions, which is used to achieve accurate positioning of the pole features and the contour features to be welded. In this example, the benchmark feature library is a structured database that stores the pole features, the contour features to be welded, and the corresponding positioning information in the benchmark image, providing a standard reference for feature verification of the prediction results. In this example, dynamic images are a series of continuous images of the welding area acquired in real time during the welding process. They completely record the characteristic changes during the welding process and serve as a data source for time-series trend analysis and anomaly identification. In this example, time-series trend analysis involves constructing a sequence of continuous dynamic images in chronological order, extracting welding area features from each frame, capturing the temporal dependencies between features, and constructing a trend curve to determine welding stability and predict welding results. In this example, the prediction results are as follows: the offset prediction range is determined based on the welding stability, and the offset results with the highest matching degree with the trend curve are selected, including the predicted pole features and the predicted welding profile features. In this example, the welding control signal is a signal used to guide the adjustment of process parameters. In this example, local parameter adjustment is the process of making targeted adjustments to the welding process parameters (such as weld joint position, welding speed, etc.) for a specific welding abnormality location based on the welding control signal. In this example, the full-process database is a database that records on-site data throughout the entire welding process. In this example, overall parameter tuning is the process of systematically optimizing and adjusting the baseline parameters of all welding processes based on anomaly patterns mined from the full-process database.

[0021] The working principle and beneficial effects of the above technical solution are as follows: To achieve real-time, accurate detection and efficient correction of welding misalignment in new energy battery terminals, firstly, by processing reference images and constructing a reference coordinate system, the influence of interference factors such as the metallic luster of the terminal is effectively eliminated on feature recognition, ensuring the accuracy of positioning of terminal features and the contour features to be welded. Simultaneously, the generated reference feature library provides a standardized reference, avoiding detection deviations caused by a lack of unified standards, and adapting to the detection needs of different models of welding plates, thus improving the method's versatility. Then, real-time acquisition of dynamic images and time-series trend analysis continuously track feature changes during the welding process, promptly capturing early signs of misalignment. The prediction results obtained through trend analysis, combined with the reference feature library, allow for rapid determination of whether the current welding status is normal, saving time for subsequent anomaly handling, preventing further expansion of defects, and ensuring the continuity of the welding process. Furthermore, when the calibration result is abnormal, the welding area is separated... By matching the abnormal feature template with non-welded areas, the scope of anomaly detection is quickly narrowed, improving detection efficiency, accurately identifying abnormal states of poles and welding contours, and locating specific welding anomaly positions. This avoids secondary damage caused by blind handling and provides accurate location data for subsequent analysis of anomaly causes. Furthermore, based on the welding offset of the anomaly position, a suitable welding control signal is generated, enabling targeted local parameter adjustment of the welding process. This avoids the impact of overall parameter adjustment on normal welding areas and improves the accuracy of correction. Finally, the construction of a full-process database enables complete retention and reuse of welding process data, providing rich data support for anomaly pattern mining. By mining the anomaly patterns in the welding area and adjusting the overall parameters accordingly, the baseline parameters of each welding process can be optimized in a targeted manner. This solves the shortcomings of traditional local correction, which cannot eradicate common problems, and significantly improves the stability and reliability of overall welding production, providing core technical support for large-scale and standardized production.

[0022] Example 2: Based on Example 1, the offset detection method for welding terminals of new energy batteries, step 1, is as follows: Figure 2 As shown, it includes: Step 11: Extract the metallic luster features contained in the reference image, input the reference image into the polarization filter mapping space, iteratively adjust the polarization direction and polarization degree of the polarization filter mapping space until the feature value of the metallic luster features is lower than a specified threshold, and then optimize the grayscale value of the mapped reference image to obtain a high-quality reference image. Step 12: Select several stable feature points contained in the high-quality reference image, randomly select a number of stable feature points to construct a feature point set, use the least squares method to fit and calculate each feature point set, determine the X-axis direction and Y-axis direction of the reference coordinate system based on the calculation results, and establish the reference coordinate system with the geometric center of the plate to be welded as the origin. Step 13: Perform edge detection on the high-quality reference image to obtain several polar features contained in the reference image, and use the active contour model algorithm to perform contour tracking on the region to be welded in the high-quality reference image to obtain the contour features to be welded. Step 14: Input each of the pole features and each of the weldable contour features into the reference coordinate system to obtain the pole positioning information and pole information entropy of the weldable plate, as well as the weld contour positioning information and weld information entropy. Derive several execution standards of the weldable plate and generate the corresponding reference feature library.

[0023] In this example, metallic luster features refer to the image features corresponding to the reflective areas of the pole in the reference image caused by the metallic material properties. In this example, the polarization filter mapping space is a virtual space constructed by combining the reference image with the optical properties of the polarization filter. It is used to simulate the image imaging effect under different polarization parameters. Within this space, the reflected light rays in a specific direction are filtered by adjusting the parameters to suppress the interference of metallic luster. In this example, polarization direction and degree of polarization: polarization direction refers to the direction of light vibration that the polarizing filter allows to pass through, and degree of polarization indicates the degree to which the polarizing filter polarizes unpolarized light; In this example, the threshold is defined as a pre-set critical value used to determine whether the metallic luster interference has been eliminated, and it is determined based on a large amount of experimental data. In this example, stable feature points refer to points on the edge or surface of the plate to be welded in a high-quality reference image that have obvious grayscale value changes, clear outlines, and are not easily affected by external interference. In this example, fitting calculation refers to the process of using mathematical methods such as the least squares method to fit the selected stable feature point set to obtain a straight line or curve model that can reflect the distribution law of the feature points. In this example, contour tracking refers to the process of continuously tracking and extracting the boundary contour along the boundary of the area to be welded in a high-quality reference image using technologies such as active contour model algorithms. In this example, polar position information refers to the specific location data of the polar pole in the coordinate system obtained after mapping the extracted polar pole features to the reference coordinate system. In this example, polar information entropy is an indicator used to measure the uncertainty or information content of polar feature data. In this example, welding contour positioning information refers to the specific position data of the contour in the coordinate system obtained after mapping the extracted features of the contour to be welded to the reference coordinate system. In this example, welding information entropy is an indicator that measures the amount of information in the feature data of the contour to be welded. In this example, the execution standard refers to the range of standard parameters derived from pole positioning information, pole information entropy, welding contour positioning information, and welding information entropy, used to determine whether the pole and welding contour are normal during the welding process.

[0024] The working principle and beneficial effects of the above technical solution are as follows: To ensure the accuracy and stability of the entire detection process from the source, firstly, by extracting metallic luster features and iteratively adjusting parameters in the polarization filter mapping space, the metallic luster feature value is made lower than a specified threshold, successfully suppressing the feature blurring problem caused by reflection from the polarimetric metal surface. Then, by optimizing the grayscale value, a high-quality reference image is obtained, providing a clear image foundation for subsequent coordinate system construction and feature extraction, avoiding subsequent detection deviations caused by poor image quality. Next, by selecting stable feature points and constructing multiple feature point sets, the least squares method is used for fitting calculation, effectively reducing the impact of single feature point errors on coordinate system construction. The coordinate axis direction is determined with the geometric center of the plate to be welded as the origin, ensuring precise matching between the coordinate system and the actual geometry of the plate to be welded, and ensuring a unified coordinate reference for subsequent feature positioning. Furthermore, the reliability lays the foundation for accurate feature mapping and comparison. Edge detection technology can efficiently and completely acquire the contour and key geometric features of the pole piece, avoiding omissions. An active contour model algorithm is used for contour tracking, accurately locking the boundary of the area to be welded, obtaining continuous and complete contour features. Finally, the extracted features are mapped to the reference coordinate system to obtain positioning information. Combined with information entropy analysis, core features with high discriminative power can be selected, reducing the interference of redundant features on subsequent calibration. By deriving execution standards, the reference feature library not only contains feature data but also has clear judgment criteria. Simultaneously, the structured feature library can adapt to the detection needs of different models of boards to be welded, improving the method's versatility. Through these methods, accurate, unified, and reliable reference support can be provided for subsequent welding offset detection, calibration, and anomaly handling.

[0025] Example 3: Based on Example 1, the offset detection method for welding terminals of new energy batteries, step 2 includes: Step 21: Obtain the welding operation progress signal of the welding area. When the progress signal triggers the start of welding, construct a frequency dynamic acquisition scheme based on the welding head movement speed signal, acquire real-time welding images of the welding head, and statistically analyze all the real-time welding images to construct a dynamic image of the welding area. Step 22: Construct a time-series image sequence from the continuously acquired dynamic images in chronological order, extract the welding area feature information contained in each dynamic image using a convolutional neural network, capture the time dependency relationship between different welding area feature information to construct the trend curve of the welding area, and determine the welding stability of the current welding process. Step 23: Based on the welding stability, determine the offset prediction range of the current welding plate to obtain several offset results, filter the prediction results with the highest matching degree with the trend curve, and identify several prediction pole features and prediction welding profile features contained in the prediction results. Step 24: Use a weighted summation method to fuse the predicted pole feature with each pole feature in the benchmark feature library, and fuse the predicted welding contour feature with each welding contour feature in the benchmark feature library to obtain the comprehensive similarity and comprehensive overlap, and generate the verification result.

[0026] In this example, process signals refer to signals generated by welding equipment during welding operations that reflect the real-time status of the welding operation. In this example, the welding start factor refers to the preset, specific conditions used to trigger the start of dynamic image acquisition; In this example, the frequency dynamic acquisition scheme refers to the acquisition strategy that adaptively adjusts the image acquisition frequency based on the welding head moving speed signal. The faster the welding head moves, the more drastic the change in the welding area per unit time. The slower the speed, the more appropriate the acquisition frequency can be reduced to save system computing resources. In this example, real-time welding image refers to a single-frame image of the welding area captured in real time by the imaging device according to the frequency requirements of the dynamic acquisition scheme during the welding operation. In this example, the time-series image sequence refers to a set of images formed by arranging continuously acquired real-time welding images in chronological order of their capture time. In this example, the time dependency refers to the correlation between the feature information of the welding area at different times in a time-series image sequence. In this example, the trend curve refers to the curve constructed based on the welding area feature information extracted from the time-series image sequence, combined with the time dependency between features, with time as the horizontal axis and feature parameters as the vertical axis. In this example, welding stability refers to the smoothness of the welding process as judged by the fluctuation of the trend curve. In this example, the offset prediction range refers to the range of possible offset degree and type that may occur in the current welded plate, determined based on welding stability.

[0027] The working principle and beneficial effects of the above technical solution are as follows: To achieve dynamic tracking and early prediction of the welding process, the acquisition is first triggered by obtaining welding operation progress signals, avoiding resource waste caused by invalid acquisition. A dynamic acquisition scheme is constructed based on the welding head movement speed, which can adaptively adjust the acquisition frequency according to the welding rhythm to ensure that key feature changes can be captured during rapid welding. Then, a time sequence is constructed from the dynamic images, and features are extracted by combining convolutional neural networks to comprehensively obtain key information of the welding area. By capturing the time dependence between features to construct a trend curve, the changing trend of the welding process can be intuitively presented, and the welding stability can be accurately judged. Furthermore, the offset prediction range is determined based on the welding stability, making the prediction more accurate. The results more closely reflect the actual welding state, avoiding invalid calculations caused by indiscriminate prediction. By selecting the prediction results with the highest matching degree with the trend curve, the most likely deviation can be accurately identified. At the same time, the predicted pole and contour features are extracted, providing a clear target for subsequent calibration with the benchmark feature library and improving calibration efficiency. Finally, a weighted summation method is used to fuse features, which can allocate weights according to the degree of influence of pole and welding contour features on welding quality, making the calculation of comprehensive similarity and comprehensive overlap more in line with actual inspection needs. The calibration results generated by dual fusion comparison can comprehensively determine whether the current welding state is normal, avoiding misjudgments caused by single feature comparison, and providing an accurate decision-making basis for subsequent anomaly handling or normal process advancement.

[0028] Example 4: Based on Example 1, the offset detection method for welding terminals of new energy batteries, step 3 includes: Step 31: When the calibration result is abnormal, identify the preliminary information of the abnormal features contained in the calibration result, and at the same time separate the welding area and non-welding area in the dynamic image. In the separated welding area image, slide to match the abnormal feature template and mark the area with a matching degree higher than the specified matching value as the suspected abnormal area. Step 32: Construct a lightweight convolutional neural network model. Input the local image of the suspected abnormal region into the extraction layer, feature enhancement layer and classification recognition layer of the lightweight convolutional neural network model for feature extraction. Determine the pole abnormality state of the current welding plate based on the output parameters of each layer. Step 33: Use the subpixel edge detection algorithm to extract edges from the local image to obtain the subpixel level precise contour of the welding contour. Then, accurately match the subpixel level precise contour with the reference features of the contour to be welded in the reference feature library to determine the abnormal state of the welding contour of the current welding plate. Step 34: Locate the first image position corresponding to each pole state and the second image position corresponding to each welding contour state in the dynamic image, input the dynamic image into the reference coordinate system to identify each image position, and determine the current welding abnormality position of the current welding plate.

[0029] In this example, the preliminary information on abnormal features includes: basic information in the calibration results that deviates from the baseline features, such as signs of pole feature offset, indications of discontinuous welding contours, and identification of parts with feature similarity below the threshold. In this example, the sliding matching anomaly feature template refers to a feature template pre-constructed based on common welding anomaly types. By setting a fixed-size sliding window, the template is moved pixel by pixel or region by region in the separated welding area image for comparison, and the matching degree between the template and the local area of ​​the image is calculated. In this example, the extraction layer is a basic functional layer in the lightweight convolutional neural network model. Its core function is to extract basic features from the local image of the suspected abnormal region. In this example, the feature enhancement layer is a functional layer located between the extraction layer and the classification and recognition layer in the lightweight convolutional neural network model. Its main function is to enhance and filter the basic features output by the extraction layer. In this example, the classification and recognition layer is the output layer in the lightweight convolutional neural network model. Its core function is to classify and judge the enhanced feature vectors. In this example, the output parameter refers to the data metrics output by the extraction layer, feature enhancement layer, and classification recognition layer of the lightweight convolutional neural network model after processing by each layer. In this example, edge extraction refers to the process of using subpixel edge detection algorithms to process local images of suspected abnormal areas, identify and extract areas in the image where grayscale values ​​change drastically. In this example, subpixel-level precise contour refers to the welding contour that breaks through the limitations of traditional pixel-level detection accuracy and achieves the precision within a pixel.

[0030] The working principle and beneficial effects of the above technical solution are as follows: To improve the efficiency of anomaly handling and ensure the controllability of welding quality, preliminary information on anomaly features is first identified to provide a clear direction for subsequent matching. Then, the welded area is separated from the non-welded area, focusing on the core analysis area and reducing interference from irrelevant background. Next, a sliding matching anomaly feature template is used to mark suspected anomaly areas, avoiding blindly checking the entire image and significantly shortening the time cost of anomaly localization, laying the foundation for subsequent refined identification. Then, a lightweight convolutional neural network model is constructed to reduce the computational complexity of the model while ensuring the accuracy of feature extraction, adapting to the needs of real-time detection. Through layered processing of extraction layer, feature enhancement layer, and classification recognition layer, the abnormal features of poles in suspected anomaly areas can be deeply mined. The abnormal state of the poles is comprehensively judged by combining the output parameters of each layer. To avoid misjudgments caused by single feature analysis and improve the accuracy of pole anomaly identification, a sub-pixel edge detection algorithm is further used to extract contours, breaking through the accuracy limitations of traditional pixel-level detection and obtaining sub-pixel-level accurate contours. This ensures the integrity of contour features and accurately matches the accurate contours with the baseline contours in the baseline feature library. This clearly identifies subtle anomalies such as contour gaps, offsets, and protrusions, clarifying the anomaly type and degree of the welding contour. This provides crucial support for the accurate location of anomalies. Finally, by locating the image positions corresponding to the pole and welding contour anomalies and mapping them to the baseline coordinate system to transform them into global coordinates, the description of anomaly positions becomes more unified and accurate. This clarifies the specific range of the current welding anomaly position, making the generation of subsequent welding control signals more targeted and ensuring that local parameter adjustments can accurately act on the anomaly area, thus improving the correction effect.

[0031] Example 5: Based on Example 4, the offset detection method for welding terminals of new energy batteries further includes: Identify the pre-welding information and post-welding information corresponding to each current welding anomaly location; The post-welding information is analyzed using the reference feature library to determine and display the cause of the welding abnormality corresponding to the current welding abnormality location.

[0032] In this example, the post-weld information is analyzed using the reference feature library to determine and display the cause of the welding abnormality corresponding to the current welding abnormality location. Specifically, this includes: Anomaly data packet construction. Visual features of the anomaly location from the image processing unit, real-time welding process parameter stream from the equipment controller, and spatiotemporal markers of the anomaly occurrence are integrated to form a time-synchronized anomaly event data packet.

[0033] The information in the data package is compared with the benchmark feature library in depth: the abnormal contour is differentially calculated with the standard contour in the library to quantify the geometric deviation; the abnormal area image is matched with the typical defect feature map in the library to identify the defect pattern; the real-time process parameters are compared with the standard process window in the library to calculate the parameter deviation.

[0034] The pre-defined expert rule base performs logical reasoning based on the extracted abnormal features. For example, if the feature satisfies {overall pole offset, and the offset direction is consistent}, the rule is triggered, and the diagnosis is "workpiece positioning failure"; if the feature satisfies {local poor weld bead formation, and the welding current is abnormal at the corresponding moment}, the diagnosis is "unstable welding energy input". The system integrates the conclusions of all triggered rules and outputs a list of potential root causes sorted by confidence level.

[0035] The working principle and beneficial effects of the above technical solution are as follows: By analyzing the causes of abnormalities during the welding process in real time, it provides reference for the staff, making it easier to quickly identify and modify abnormalities.

[0036] Example 6: Based on Example 1, the offset detection method for welding terminals of new energy batteries, step 4 includes: Step 41: Extract the pole column abnormality state parameters and welding profile abnormality state parameters corresponding to each current welding abnormality position, and construct the pole column offset vector and profile offset vector of the current welding abnormality position by combining the pole column offset direction and welding profile offset direction corresponding to the current welding position. Step 42: Determine the parameter adjustment direction and parameter adjustment range of the current welding abnormality position based on the pole offset vector and the contour offset vector, and perform PID adjustment on the parameter adjustment direction and parameter adjustment range corresponding to the same current welding plate to obtain the optimal parameter adjustment scheme corresponding to each current welding abnormality position; Step 43: Generate a standardized digital signal corresponding to the current welding anomaly position according to the optimal parameter adjustment scheme, perform anti-interference processing on the standardized digital signal to obtain the welding control signal corresponding to each current welding position, and transmit it to the welding area; Step 44: Locate the welding equipment corresponding to the current welding plate in the welding area, and use the welding control signal to locally adjust the welding process of the welding equipment.

[0037] In this example, the pole abnormality status parameter refers to the specific data indicators used to quantify the degree of pole abnormality at the current welding abnormality location. In this example, the welding profile abnormality status parameter refers to a specific data indicator used to quantify the degree of welding profile abnormality at the current welding abnormality location. In this example, the pole offset vector refers to a two-dimensional vector constructed by combining the pole abnormal state parameters and the pole offset direction. The direction of the vector corresponds to the actual direction of the pole offset, and the magnitude of the vector corresponds to the magnitude of the pole offset. In this example, the profile offset vector is a two-dimensional vector constructed by combining the welding profile anomaly state parameters and the welding profile offset direction. The direction of the vector corresponds to the orientation of the welding profile offset or defect, and the magnitude of the vector corresponds to the quantification degree of the profile anomaly. In this example, the parameter adjustment direction refers to the specific direction in which the welding process parameters need to be adjusted, as determined by the pole offset vector and the profile offset vector. In this example, the parameter adjustment range refers to the range of adjustment of the welding process parameters determined by the modulus of the pole offset vector and the contour offset vector. In this example, PID control refers to a closed-loop control method based on proportional (P), integral (I), and derivative (D), which is used to integrate the parameter adjustment direction and range of multiple abnormal positions on the same welding plate. In this example, the optimal parameter tuning scheme refers to the welding process parameter adjustment scheme formulated for each current welding abnormality position after PID tuning and optimization, which can accurately and stably correct the abnormality. In this example, standardized digital signal refers to the conversion of adjustment parameters and control commands in the optimal parameter tuning scheme into digital coded signals that conform to industrial control standards. In this example, anti-interference processing refers to the process of protecting standardized digital signals from interference factors such as electromagnetic interference, vibration interference, and voltage fluctuations present at the welding site.

[0038] The working principle and beneficial effects of the above technical solution are as follows: In order to ensure the continuity of the welding process, avoid production interruption caused by abnormal handling, and reduce the defective product rate and production cost, the abnormal state parameters of the pole and welding profile are extracted first, and the offset vector is constructed by combining the offset direction. This realizes the standardized and quantitative expression of abnormal information, avoids the parameter adjustment deviation caused by vague qualitative description, and makes the determination of the subsequent parameter adjustment direction and range more scientific and targeted. Then, by integrating and optimizing the parameter adjustment requirements of the same welding plate through PID control, interference between parameters can be effectively eliminated, over-adjustment or under-adjustment can be avoided, the optimality of the parameter adjustment scheme can be guaranteed, and the stability of the correction effect can be improved. Furthermore, the optimal parameter adjustment scheme is transformed into a standardized digital signal, realizing the standardized expression of parameter adjustment information, which is convenient for welding equipment to identify. Through anti-interference processing, the influence of electromagnetic, vibration and other interference factors on the signal at the welding site is reduced, avoiding parameter adjustment errors caused by signal distortion. Finally, by locating the welding equipment corresponding to the current welding plate, it is ensured that the control signal is accurately applied to the target equipment, and local parameter adjustment is performed for abnormal positions, avoiding unnecessary interference to the normal welding area. The abnormal correction is completed without interrupting the overall welding process, which not only improves the correction efficiency, but also ensures the overall welding quality.

[0039] Example 7: Based on Example 1, the offset detection method for welding terminals of new energy batteries, step 5 includes: Step 51: Collect the offset data, control parameters and imaging information of the entire welding process and perform standardization processing respectively to build a full process database. Locate the full data corresponding to each local parameter adjustment process in the full process database, and set corresponding reason labels for the full data based on the parameter adjustment reason corresponding to each local parameter adjustment. Step 52: Mine several related data corresponding to each full data in the full process database, and logically sort out the related data corresponding to the full data with the same cause label to obtain the abnormal cause corresponding to each parameter tuning cause; Step 53: Simulate each of the above-mentioned abnormal causes under different welding scenarios, determine the abnormal patterns corresponding to each of the above-mentioned abnormal causes, analyze the abnormal manifestations of the abnormal patterns on each of the above-mentioned welding processes, and optimize each of the above-mentioned abnormal manifestations. Step 54: Based on the optimization results, confirm the content to be optimized for each welding process, and perform overall parameter adjustment for each welding process.

[0040] In this example, offset data refers to various quantitative data related to welding offset throughout the entire welding process, including coordinate data of abnormal positions, offset between pole and welding profile, magnitude and direction of offset vector, and time node of offset occurrence. In this example, the control parameters refer to various welding process parameters and control parameters that are adjusted during the welding process to correct the offset, including the spatial position adjustment value of the welding head, the welding current / voltage adjustment value, the welding speed adjustment value, the welding shielding gas flow rate adjustment value, and the coding parameters of the welding control signal, etc. In this example, image information refers to various types of image data collected throughout the welding process, including the reference image of the plate to be welded, dynamic images during the welding process, local images of suspected abnormal areas, and comparison images before and after parameter adjustment. In this example, "full data" refers to a complete dataset formed by integrating the corresponding offset data, control parameters, imaging information, welding status data before and after parameter adjustment, and production environment data for each local parameter adjustment process. In this example, the reason label refers to the classification label added to each set of full data based on the specific reason for the local parameter tuning. In this example, related data refers to data in the entire process database that has a relationship with a certain set of data. In this example, logical sorting refers to the systematic organization and analysis of all data with the same cause label and its related data; In this example, different welding scenarios refer to various differentiated production conditions that may occur during the welding production process, mainly including different welding equipment models, different materials of the plates to be welded, different production environments, and different welding process types. In this example, the abnormal manifestation refers to the abnormal pattern corresponding to a certain abnormal cause, and the specific problems or phenomena that appear in a specific welding process.

[0041] The working principle and beneficial effects of the above technical solution are as follows: To improve the stability and standardization of welding production and provide sustainable technical support for large-scale production, the solution firstly collects and standardizes multi-type data throughout the entire process, eliminating analytical obstacles caused by data format differences. Cause labels are then set for local parameter tuning data, making subsequent data mining more targeted and ensuring data traceability, facilitating the tracing of anomalies. Next, by mining related data from the entire dataset and integrating data with the same cause labels for logical analysis, the core anomaly causes directly related to parameter tuning can be extracted from the massive dataset, avoiding the blind elimination of anomaly causes. The investigation allows subsequent optimization work to target core issues, improve optimization efficiency, and further simulate the causes of anomalies under different welding scenarios to ensure that the derived anomaly patterns have broad applicability and avoid the limitations of patterns in a single scenario. By analyzing the anomaly patterns and their manifestation in the welding process, the weak points in the process links can be clearly located. Finally, based on the optimization results, the content to be optimized for each process is determined, avoiding the blindness of overall parameter tuning. Performing overall parameter tuning for each welding process separately can solve common anomaly problems from the root, significantly improve the stability and reliability of the overall welding process, reduce the probability of subsequent deviation defects, and improve production efficiency and product qualification rate.

[0042] Example 8: Based on Example 4, the offset detection method for welding terminals of new energy batteries further includes: When the calibration result is normal, obtain the current process being executed for the current welding plate; Add a recommendation weight to the currently executed process; The recommended process is sequentially recommended to each welding area based on the recommended weight.

[0043] The working principle and beneficial effects of the above technical solution are as follows: It can prioritize the selection of mature processes with stable welding quality and strong adaptability, avoiding the inefficiency and increased costs caused by repeatedly exploring the optimal process in each welding area. It can also rapidly improve the uniformity and stability of the welding process of the entire production line through standardization and promotion, further reducing the overall deviation defect rate, while shortening the process debugging cycle of new welding areas, thus helping to improve overall production efficiency and product qualification rate.

[0044] Example 9: Based on Example 6, the offset detection method for welding terminals of new energy batteries further includes: When the magnitude of the first vector corresponding to the pole offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. When the second vector magnitude corresponding to the contour offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. The defective products are then transported to the scrap area to await destruction.

[0045] The working principle and beneficial effects of the above technical solution are as follows: timely diversion of defective products to the scrap area can prevent defective products from entering subsequent processing stages, thus avoiding additional processing costs and resource waste. At the same time, it can ensure the overall quality level of finished products, reduce quality risks and after-sales costs caused by the outflow of unqualified products, and improve the standardization and efficiency of the production process.

[0046] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for offset detection of new energy battery pole welding, characterized in that, include: Step 1: Construct a reference coordinate system based on the reference image of the board to be welded, input the polar features and the contour features to be welded contained in the reference image into the reference coordinate system for feature localization, and generate the reference feature library of the board to be welded. Step 2: Real-time acquisition of dynamic images of the welding area, time-series trend analysis of the dynamic images to obtain the prediction results of the current welding plate, and feature verification of the prediction results using the benchmark feature library to obtain the verification results; Step 3: When the calibration result is abnormal, identify the abnormal state of the pole post and the abnormal state of the welding contour of the current welding plate in the dynamic image, and determine the current welding abnormality position corresponding to the current welding plate. Step 4: Generate a welding control signal based on the welding offset corresponding to each current welding abnormality position, and perform local parameter adjustment on the welding process of the current welding plate; Step 5: Record on-site data throughout the welding process to build a full-process database. In the full-process database, discover abnormal patterns in the welding area and use these abnormal patterns to adjust the parameters of all welding processes as a whole.

2. The offset detection method for welding terminals of new energy batteries as described in claim 1, characterized in that, Step 1 includes: Step 11: Extract the metallic luster features contained in the reference image, input the reference image into the polarization filter mapping space, iteratively adjust the polarization direction and polarization degree of the polarization filter mapping space until the feature value of the metallic luster features is lower than a specified threshold, and then optimize the grayscale value of the mapped reference image to obtain a high-quality reference image. Step 12: Select several stable feature points contained in the high-quality reference image, randomly select a number of stable feature points to construct a feature point set, use the least squares method to fit and calculate each feature point set, determine the X-axis direction and Y-axis direction of the reference coordinate system based on the calculation results, and establish the reference coordinate system with the geometric center of the plate to be welded as the origin. Step 13: Perform edge detection on the high-quality reference image to obtain several polar features contained in the reference image, and use the active contour model algorithm to perform contour tracking on the region to be welded in the high-quality reference image to obtain the contour features to be welded. Step 14: Input each of the pole features and each of the weldable contour features into the reference coordinate system to obtain the pole positioning information and pole information entropy of the weldable plate, as well as the weld contour positioning information and weld information entropy. Derive several execution standards of the weldable plate and generate the corresponding reference feature library.

3. The offset detection method for welding terminals of new energy batteries as described in claim 1, characterized in that, Step 2 includes: Step 21: Obtain the welding operation progress signal of the welding area. When the progress signal triggers the start of welding, construct a frequency dynamic acquisition scheme based on the welding head movement speed signal, acquire real-time welding images of the welding head, and statistically analyze all the real-time welding images to construct a dynamic image of the welding area. Step 22: Construct a time-series image sequence from the continuously acquired dynamic images in chronological order, extract the welding area feature information contained in each dynamic image using a convolutional neural network, capture the time dependency relationship between different welding area feature information to construct the trend curve of the welding area, and determine the welding stability of the current welding process. Step 23: Based on the welding stability, determine the offset prediction range of the current welding plate to obtain several offset results, filter the prediction results with the highest matching degree with the trend curve, and identify several prediction pole features and prediction welding profile features contained in the prediction results. Step 24: Use a weighted summation method to fuse the predicted pole feature with each pole feature in the benchmark feature library, and fuse the predicted welding contour feature with each welding contour feature in the benchmark feature library to obtain the comprehensive similarity and comprehensive overlap, and generate the verification result.

4. The offset detection method for welding terminals of new energy batteries as described in claim 1, characterized in that, Step 3 includes: Step 31: When the calibration result is abnormal, identify the preliminary information of the abnormal features contained in the calibration result, and at the same time separate the welding area and non-welding area in the dynamic image. In the separated welding area image, slide to match the abnormal feature template and mark the area with a matching degree higher than the specified matching value as the suspected abnormal area. Step 32: Construct a lightweight convolutional neural network model. Input the local image of the suspected abnormal region into the extraction layer, feature enhancement layer and classification recognition layer of the lightweight convolutional neural network model for feature extraction. Determine the pole abnormality state of the current welding plate based on the output parameters of each layer. Step 33: Use the subpixel edge detection algorithm to extract edges from the local image to obtain the subpixel level precise contour of the welding contour. Then, accurately match the subpixel level precise contour with the reference features of the contour to be welded in the reference feature library to determine the abnormal state of the welding contour of the current welding plate. Step 34: Locate the first image position corresponding to each pole state and the second image position corresponding to each welding contour state in the dynamic image, input the dynamic image into the reference coordinate system to identify each image position, and determine the current welding abnormality position of the current welding plate.

5. The offset detection method for welding terminals of new energy batteries as described in claim 4, characterized in that, Also includes: Identify the pre-welding information and post-welding information corresponding to each current welding anomaly location; The post-welding information is analyzed using the reference feature library to determine and display the cause of the welding abnormality corresponding to the current welding abnormality location.

6. The offset detection method for welding terminals of new energy batteries as described in claim 1, characterized in that, Step 4 includes: Step 41: Extract the pole column abnormality state parameters and welding profile abnormality state parameters corresponding to each current welding abnormality position, and construct the pole column offset vector and profile offset vector of the current welding abnormality position by combining the pole column offset direction and welding profile offset direction corresponding to the current welding position. Step 42: Determine the parameter adjustment direction and parameter adjustment range of the current welding abnormality position based on the pole offset vector and the contour offset vector, and perform PID adjustment on the parameter adjustment direction and parameter adjustment range corresponding to the same current welding plate to obtain the optimal parameter adjustment scheme corresponding to each current welding abnormality position; Step 43: Generate a standardized digital signal corresponding to the current welding anomaly position according to the optimal parameter adjustment scheme, perform anti-interference processing on the standardized digital signal to obtain the welding control signal corresponding to each current welding position, and transmit it to the welding area; Step 44: Locate the welding equipment corresponding to the current welding plate in the welding area, and use the welding control signal to locally adjust the welding process of the welding equipment.

7. The offset detection method for welding terminals of new energy batteries as described in claim 1, characterized in that, Step 5 includes: Step 51: Collect the offset data, control parameters and imaging information of the entire welding process and perform standardization processing respectively to build a full process database. Locate the full data corresponding to each local parameter adjustment process in the full process database, and set corresponding reason labels for the full data based on the parameter adjustment reason corresponding to each local parameter adjustment. Step 52: Mine several related data corresponding to each full data in the full process database, and logically sort out the related data corresponding to the full data with the same cause label to obtain the abnormal cause corresponding to each parameter tuning cause; Step 53: Simulate each of the above-mentioned abnormal causes under different welding scenarios, determine the abnormal patterns corresponding to each of the above-mentioned abnormal causes, analyze the abnormal manifestations of the abnormal patterns on each of the above-mentioned welding processes, and optimize each of the above-mentioned abnormal manifestations. Step 54: Based on the optimization results, confirm the content to be optimized for each welding process, and perform overall parameter adjustment for each welding process.

8. The offset detection method for welding terminals of new energy batteries as described in claim 4, characterized in that, Also includes: When the calibration result is normal, obtain the current process being executed for the current welding plate; Add a recommendation weight to the currently executed process; The recommended process is sequentially recommended to each welding area based on the recommended weight.

9. The offset detection method for welding terminals of new energy batteries as described in claim 6, characterized in that, Also includes: When the magnitude of the first vector corresponding to the pole offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. When the second vector magnitude corresponding to the contour offset vector is greater than a specified magnitude threshold, the current welding plate is determined to be a defective product. The defective products are then transported to the scrap area to await destruction.

Images