Self-adaptive fusion depth control system for TIG (Tungsten Inert Gas) welding of coiled pipe

Abstract

The invention discloses a TIG (Tungsten Inert Gas) welding penetration depth self-adaptive control system for a coiled pipe, belongs to the technical field of welding automatic control, and aims to solve the problems of inconsistent penetration depth control and real-time early warning of abnormal processing in TIG welding of the coiled pipe. The method comprises the steps that welding parameters are initialized based on parameters such as the material and the pipe diameter of the coiled pipe, and the target fusion depth is determined; identifying a straight line segment and a curve segment through the panoramic image, marking a target welding point and generating a welding path; determining a local processing space according to welding point position division types and differentiation, and verifying and updating; positioning precursor, current and subsequent points during updating, collecting images to extract multi-modal features, and adaptively updating the target fusion depth based on the geometric change of the subsequent points and the quality features of the precursor points; and during welding, processing quality and intervention abnormity early warning is carried out according to the real-time multi-mode characteristics, the processing parameters and the environment parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of welding automation control, and more particularly to a serpentine tube TIG welding penetration adaptive control system. Background Art

[0002] In high-end equipment manufacturing, coiled tubes serve as critical components for heat exchange and fluid transport, and their welding quality plays a crucial role in equipment performance and safety. TIG welding, with its advantages of arc stability and aesthetically pleasing weld seams, has become a common process for coiled tube welding. However, the complex structure of coiled tubes, characterized by numerous alternating curved and straight sections, varying diameters, and sudden changes in curvature, presents numerous challenges for traditional welding technologies. Firstly, it is difficult to precisely control the depth of penetration across different areas, resulting in uneven weld strength and a high risk of over- or under-welding. Secondly, existing monitoring systems, often based on single parameters or simple thresholds, fail to detect potential anomalies during the welding process. Consequently, anomaly warnings are delayed, often resulting in batches of defective products by the time defects are discovered. Furthermore, manual welding path planning is inefficient and difficult to adapt to the complex spatial layout of coiled tubes. Frequent adjustments to the equipment's posture not only reduce production efficiency but also increase the risk of collisions. Furthermore, traditional welding systems lack the ability to dynamically adjust to fluctuations in material properties or changes in equipment status, making it difficult to ensure consistent weld quality. This severely restricts the efficiency and quality of high-end equipment manufacturing. Therefore, in order to overcome these limitations, the present invention proposes a serpentine tube TIG welding penetration adaptive control system. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a serpentine tube TIG welding penetration adaptive control system, which solves the problem of inconsistent penetration control caused by changes in workpiece geometric characteristics, heat conduction differences and process fluctuations in serpentine tube TIG welding, as well as the real-time early warning problem of abnormal processing quality and intervention anomalies during welding.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A serpentine tube TIG welding penetration adaptive control system, comprising:

[0006] Initialize the initial processing parameters of the TIG welding equipment according to the basic parameters of the serpentine tube and determine the target penetration value;

[0007] Capture panoramic images of the TIG welding equipment's working area, identify the straight and curved segments of the serpentine tube, and, based on a welding point template, traverse the image sub-regions of the panoramic image to mark the target welding points. Then, construct a weighted directed graph with the target welding points as nodes, perform path search, and generate a welding path.

[0008] The target weld points are classified into different types according to their positions on the serpentine tube, and the local processing space size is determined by differentiation. The predecessor and successor points of the target weld points are defined based on the welding path. The local processing space size is verified and updated to obtain the local processing space of the target weld points.

[0009] When the local processing space is updated, the predecessor point, target welding point, and successor point in the local processing space are located, and images within the welding window are collected respectively. The multimodal features in the local processing space are extracted, and the target penetration depth is adaptively updated:

[0010] Based on the geometric change characteristics of the subsequent point, the initial compensation coefficient is corrected to update its target penetration depth. The target penetration depth of the initial target welding point is updated according to the local processing space. Then, the welding status is judged based on the quality characteristic parameters of the predecessor point and the current target penetration depth is corrected.

[0011] When the welding process is driven by the target penetration value, processing quality abnormality warning and intervention abnormality warning are carried out based on the multi-modal characteristics, processing parameters and environmental parameters collected in real time.

[0012] Specifically, the specific steps of adaptively updating the target penetration depth include:

[0013] The multimodal features in the local processing space of the target welding point are sequentially obtained, and according to the geometric change features of the subsequent points of the target welding point, a preset geometric feature change threshold is used to determine whether to perform pre-compensation processing on the subsequent points;

[0014] If so, the initial compensation coefficients of the geometric change features are weightedly modified according to the ratio of the geometric change feature to the corresponding geometric change threshold value, and used as the compensation coefficients of the geometric change features;

[0015] The compensation coefficients of each geometric change feature are weighted and integrated to calculate the comprehensive compensation coefficient. The target penetration value of the current target welding point is multiplied by the comprehensive compensation coefficient to obtain the initial target penetration value of the subsequent point.

[0016] If not, the target penetration value of the target welding point is used as the initial target penetration value of the subsequent point.

[0017] Specifically, the specific steps of adaptively updating the target penetration depth also include:

[0018] After the welding operation of the target welding point in the local processing space is completed, the local processing space is updated, and the initial target penetration value of the target welding point in the current local processing space is set according to the initial target penetration value of the successor point in the previous local processing space;

[0019] Extract the quality characteristic parameters of the precursor points, including weld width, weld height, contour gradient matrix, and weld contour void length; use the quality assessment model to determine the welding quality status, including qualified welding, excessive welding, and missing welding, and analyze the contribution of each quality characteristic parameter to the classification result through the SHAP value;

[0020] The quality assessment model is used to classify and identify the welding quality status based on the nonlinear mapping relationship of weld quality characteristic parameters, and is obtained through training and fitting based on historical welding data;

[0021] If the quality assessment model determines that the welding is qualified, the initial target penetration value of the current target weld point is used as the target penetration value of the welding process; if the quality assessment model determines that the welding is excessive or missing, the dominant quality feature is determined according to the SHAP value, and the penetration nonlinear mapping function is called to calculate the penetration correction coefficient according to the direction and size of the dominant quality feature deviation vector, and the initial target penetration value of the target weld point is corrected to serve as the target penetration value of the welding process; the penetration nonlinear mapping function is used to map the deviation vector of the dominant quality feature to the penetration correction coefficient, which is obtained through training and fitting based on historical welding data.

[0022] Specifically, the specific steps of local processing space of the target welding point include:

[0023] Generate an ordered sequence of target welding points based on the welding path, establish spatial topological associations between the target welding points, and define predecessor points and successor points for each target welding point according to the direction of the welding path;

[0024] The size of the local processing space is determined based on the type of welding point, namely:

[0025] The initial local processing space size of the straight welding point is determined by the length of the straight section of the serpentine tube where it is located and the diameter of the serpentine tube;

[0026] The initial local processing space size of the curved welding point is determined by the length of the curved section of the serpentine tube, the diameter of the serpentine tube and the local curvature characteristics.

[0027] Taking the target welding point as the center, the initial local processing space is delineated according to the size of the initial local processing space, and the initial local processing space is verified to verify whether the initial local processing space contains the predecessor point and the successor point of the target welding point;

[0028] If it is included, the initial local processing space is determined as the local processing space;

[0029] If the predecessor point is missing, the predecessor point of the target welding point is located according to the welding path and is included in the initial local processing space. If the successor point is missing, the successor point of the target welding point is located according to the welding path and is included in the initial local processing space.

[0030] The initial local processing space size is updated to obtain the local processing space of the target welding point.

[0031] Specifically, the steps for generating a welding path include:

[0032] Capture panoramic images of the serpentine tube within the workspace of the TIG welding equipment, extract edge features, and construct the serpentine tube spatial coordinate system;

[0033] Perform straight line detection on edge features, identify the straight and curved segments of the serpentine tube, and calculate the curvature parameters of the curved segments, including the curvature radius and curvature change rate of the curved segments. The curvature mutation points are marked by the preset curvature mutation threshold.

[0034] Based on the welding point template image, the image sub-region is traversed through a sliding window in the panoramic image, and the normalized cross-correlation coefficient is used to calculate the similarity between the welding point template and the image sub-region. If the similarity is greater than the preset similarity threshold, it is marked as a target welding point.

[0035] Specifically, the specific steps of generating the welding path also include:

[0036] The coordinate position of the target welding point is located according to the spatial coordinate system of the serpentine tube, and the target welding point is classified into a straight welding point or a curved welding point according to the position distribution of the target welding point in the straight segment and the curved segment of the serpentine tube;

[0037] Based on the classification results of welding points, a weighted directed graph with target welding points as nodes is constructed. The edge weights are determined by the Euclidean distance between target welding points, the change in the posture angle of the TIG welding equipment, and the welding accessibility.

[0038] A path search is performed on the weighted directed graph with the optimization goal of minimizing the total weight of the path. The initial welding path is generated through selection, crossover and mutation operations, and the welding path of the TIG welding equipment is obtained through smoothing.

[0039] Specifically, the specific steps of extracting multimodal features in the local processing space include:

[0040] When the local processing space is updated, the predecessor point in the current local processing space is located, the welding window is set, and the welded image in the predecessor point welding window is obtained with the predecessor point as the center;

[0041] Pre-process the welded image to separate the weld area from the base material area;

[0042] For the weld area, the weld contour is extracted to calculate the weld width, and the weld centerline is extracted to calculate the cross-sectional contour perpendicular to the centerline, and the weld residual height is fitted; the grayscale gradient amplitude of the weld edge is calculated according to the weld contour to obtain the contour gradient matrix; the length of the weld contour cavity is identified through morphological operations.

[0043] Specifically, the specific steps of extracting multimodal features in the local processing space also include:

[0044] Locate the target welding point and the subsequent point in the current local processing space, take the target welding point and the subsequent point as the center, and obtain the image to be welded in the welding window of the target welding point and the subsequent point;

[0045] After preprocessing the welding image, the geometric features of the target welding point and the subsequent points are extracted respectively. The geometric features of the target welding point and the subsequent points are compared, and the geometric feature change rate of the subsequent point relative to the target welding point is extracted to construct the geometric change feature of the subsequent point.

[0046] When processing the target welding point in the local processing space, the molten pool image in the welding window of the target welding point is dynamically captured through the synchronous trigger circuit with the target welding point as the center;

[0047] The melt pool image is segmented, the melt pool contour is extracted, the melt pool length, melt pool width and melt pool area are calculated, and the melt pool area change rate of the melt pool contour in the melt pool images of adjacent frames is calculated.

[0048] Specifically, the specific steps for early warning of abnormal processing quality include:

[0049] Obtaining melt pool features of the melt pool image of the target weld point, the melt pool features including melt pool length, melt pool width, and melt pool area change rate; and associating the melt pool features with the position of the target weld point;

[0050] Retrieve the rated processing parameters corresponding to the target penetration value, calculate the deviation between the real-time processing parameters and the rated processing parameters, and trigger a processing parameter abnormality flag if the deviation is greater than the preset processing deviation threshold;

[0051] Set the melt pool characteristic threshold, compare the melt pool characteristics of the target welding point with the corresponding melt pool characteristic threshold, and mark the melt pool abnormality;

[0052] According to the abnormal processing parameter marks and the abnormal molten pool marks, the abnormal target welding points are located, and the processing quality abnormality warning is issued according to the position distribution of the abnormal target welding points. The number of abnormal target welding points in the workspace is counted. If it is greater than the preset abnormal point threshold, a processing quality abnormality warning is issued.

[0053] Specifically, the specific steps for intervening in abnormal warnings include:

[0054] The welding quality status of the predecessor point in the current local processing space is obtained. If it is determined that there is excessive welding or missing welding, a sliding window is established based on the temporal correlation of the local processing space to perform defect repeatability analysis. If the number of times the same welding quality status appears in the sliding window exceeds the preset defect threshold, and the confidence level of a single defect is greater than the configured reliability threshold, it is marked as a recurring quality defect.

[0055] Analyze the target melting depth correction amount of the subsequent point. If it is greater than the preset correction threshold, it is marked as the correction amount exceeding the threshold; and perform correction detection, configure the correction monitoring interval, and count the number of times the correction amount exceeds the threshold within the correction monitoring interval. If it is greater than the correction number threshold, an intervention abnormality warning is issued; and when the repeated appearance mark of the predecessor point quality defect in the local processing space and the subsequent point correction amount exceeding the threshold mark are triggered at the same time, an intervention abnormality warning is issued.

[0056] Beneficial effects of the present invention:

[0057] The present invention realizes differentiated penetration control of the straight and curved sections of the serpentine tube through the dual mechanisms of geometric feature pre-compensation and quality feedback correction, significantly improving the penetration consistency and reducing welding defects; the intelligent path planning algorithm based on the weighted directed graph optimizes the welding sequence while reducing the time consumption and collision risk of equipment posture adjustment, thereby improving the overall welding efficiency; the dynamic local space management mechanism defines the processing space according to the type of welding point, ensures the integrity of multimodal feature extraction, and effectively copes with complex structural constraints; through the dual real-time monitoring of the molten pool characteristics and parameter deviations and the sliding window accumulation correction detection, rapid response to processing quality anomalies and prediction of systematic deviations are achieved, thereby improving the defect detection rate; the self-learning model based on historical data training supports the system to adaptively adjust the control parameters, and continuously improves the penetration control accuracy as data accumulates; the intervention abnormality warning mechanism can automatically stop the abnormal welding process, reduce equipment loss and quality rework costs, and form a full-process closed-loop optimization from planning, execution to monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a schematic diagram of the structure of a serpentine tube TIG welding penetration adaptive control system of the present invention;

[0059] Figure 2 A flowchart of the specific steps for generating a welding path for the present invention;

[0060] Figure 3 A flow chart for dividing the local processing space of the present invention;

[0061] Figure 4 A flowchart of specific steps for adaptively updating target penetration depth according to the present invention;

[0062] Figure 5The figure is a flow chart of the specific steps of intervening in abnormal warning according to the present invention. DETAILED DESCRIPTION

[0063] Example 1

[0064] See also Figure 1 ,This embodiment introduces a serpentine tube TIG welding penetration adaptive control system, including a parameter initialization module, a path planning module, a local feature extraction module, a penetration decision module and a risk warning module;

[0065] Before welding begins, the parameter initialization module automatically retrieves the corresponding processing parameter data from the welding process parameter database based on the coiled tube's material properties, diameter, and wall thickness specifications. This module then sets the initial processing parameters for the TIG welding equipment, including welding current, arc voltage, welding speed, shielding gas flow, tungsten electrode diameter, and pulse frequency. Furthermore, the module combines material properties with welding standards to determine the target penetration value and constructs a correlation model between penetration and processing parameters, providing benchmark data support for subsequent welding processes.

[0066] In this embodiment, basic parameters of the serpentine pipe are received, including material properties such as 304 stainless steel, carbon steel Q235, alloy steel P91, pipe diameter, and wall thickness specifications. A built-in welding process parameter database is stored and categorized by material, pipe diameter, and wall thickness. After receiving the basic parameters of the serpentine pipe, a multi-condition matching algorithm is used to retrieve corresponding data from the welding process parameter database. The process parameter data retrieved from the welding process parameter database is communicated with the TIG welding equipment via an industrial communication protocol, and parameters such as welding current, arc voltage, welding speed, shielding gas flow rate, tungsten electrode diameter, and pulse frequency are sent to the equipment controller. Based on the material properties of the serpentine pipe, such as melting point, thermal conductivity, and welding standards, a target penetration value is determined using a combination of empirical formulas and database queries. Once the target penetration is determined, it is associated with parameters such as pipe diameter and wall thickness and stored for easy subsequent access. A model correlating penetration and processing parameters is constructed using multivariate linear regression or machine learning algorithms based on historical welding data. The process parameters such as welding current, arc voltage, and welding speed are used as input variables, and the penetration depth is used as the output variable.

[0067] The path planning module is used to capture panoramic images of the serpentine tube within the TIG welding equipment's workspace, identify the coordinates and distribution of preset target weld points, and automatically distinguish between straight and curved segments of the serpentine tube, classifying the welds into straight and curved segments. Based on the weld point type and distribution, a path optimization algorithm is used to generate a welding path, ensuring a smooth, non-redundant weld trajectory while meeting the welding requirements of the complex serpentine tube structure. Based on the weld point type and spatial distribution, the module dynamically divides the local processing space for the current weld point into sliding updates in real time. The local processing space contains at least one processed weld point, one current weld point, and at least one unprocessed weld point.

[0068] In this embodiment, the path planning module uses an industrial camera to capture a panoramic image of the serpentine tube within the workspace. After preprocessing through grayscaling and filtering, it identifies preset target weld points, extracts straight line segments and fits curve segments, and categorizes the weld points into straight and curved segments. Based on the weld point type and distribution, a welding path is generated, ensuring that the welding gun motion complies with posture and obstacle avoidance constraints. Furthermore, the local processing space is divided according to the weld point type and welding path: straight line segments have a fixed range centered on the current weld point, while curved segments have a range that adjusts based on the curvature radius. Each space contains at least one processed point, one current point, and at least one unprocessed point, supporting real-time sliding updates of the space.

[0069] See also Figure 2 Preferably, the specific steps of generating the welding path include:

[0070] Before the welding operation starts, a panoramic image of the serpentine tube in the workspace of the TIG welding equipment is collected. After Gaussian filtering and histogram equalization are performed on the panoramic image to reduce noise and enhance contrast, the Canny operator is used to extract edge features to identify key geometric structures such as straight segments, curved segments, and tube diameter changes. The camera's internal and external parameters are calibrated using the Zhang Zhengyou calibration method to construct the serpentine tube's spatial coordinate system.

[0071] Hough transform is used to detect straight lines on edge features and identify the straight and curved segments of the serpentine tube. The B-spline curve is fitted using the RANSAC algorithm to calculate the curvature parameters of the curve segment, including the curvature radius and curvature change rate of the curve segment. The curvature mutation point is marked by a preset curvature mutation threshold, providing a basis for speed and posture adjustment in subsequent path planning.

[0072] Obtain the serpentine tube welding point template, traverse the image subregion in the panoramic image through a sliding window based on the welding point template image, and use the normalized cross-correlation coefficient to calculate the similarity between the welding point template and the image subregion. If the similarity is greater than the preset similarity threshold, it is marked as the target welding point;

[0073] By traversing the panoramic image of the serpentine tube, the target welding points preset on the surface of the serpentine tube are identified. The coordinate position of the target welding points is located according to the spatial coordinate system of the serpentine tube. The target welding points are classified as straight welding points or curved welding points based on their position distribution in the straight and curved segments of the serpentine tube.

[0074] Based on the weld point classification results, a weighted directed graph with the target weld points as nodes is constructed. The edge weights are determined by the Euclidean distance between the target weld points, the change in the TIG welding equipment's posture angle, and the weld accessibility. The Euclidean distance between the target weld points is used to measure the path length and control idle travel and energy consumption during welding. The change in the TIG welding equipment's posture angle is used to assess the smoothness of the welding gun's motion and avoid sudden changes in posture that lead to arc instability and molten pool fluctuations. It is determined by the difference in the angle between the welding gun axis and the surface normal of the serpentine tube at adjacent weld points. The weld accessibility is used to determine whether the welding gun can safely reach the target weld point and is determined by a collision detection algorithm based on a hierarchical bounding box.

[0075] A genetic algorithm is used to search for paths in a weighted directed graph, with the optimization goal of minimizing the total path weight. An initial welding path is generated through selection, crossover, and mutation operations. This initial welding path is then smoothed using an interpolation algorithm to obtain the welding path for the TIG welding equipment. This ensures a smooth and continuous trajectory for the welding gun between adjacent target weld points. The gun's speed is correlated with the curvature of the welding path, automatically reducing the welding speed at sudden changes in curvature and along curved sections to ensure consistent welding quality.

[0076] See also Figure 3 Preferably, the specific steps of dividing the local processing space according to the welding point type and the welding path include:

[0077] Generate an ordered sequence of target welding points based on the welding path, establish a spatial topological association between the target welding points, and define a predecessor point and a successor point for each target welding point according to the direction of the welding path. The predecessor point is used to represent the processed welding point, and the successor point is used to represent the welding point to be processed.

[0078] The size of the local processing space is determined differently according to the type of welding point. For straight welding points, the initial local processing space size is determined by the length of the straight segment of the serpentine tube and the diameter of the serpentine tube. The longer the straight segment, the larger the local processing space to cover more welding areas; the larger the diameter, the more the space range will be moderately expanded to ensure the stability of the welding operation and control the penetration depth.

[0079] For curved welds, the initial local processing space size is determined by the length of the serpentine curve segment, the diameter of the serpentine tube, and the local curvature characteristics. The curve segment length affects the overall spatial coverage; longer lengths result in a larger space. The diameter also serves as a basic parameter, influencing the basic expansion of the space. The local curvature characteristics are the core influencing factor. The greater the curvature, the smaller the local processing space size is, in order to achieve precise welding control and penetration adjustment, so as to more accurately capture the impact of curve changes on welding.

[0080] After the size calculation is completed, the initial local processing space is delineated based on the initial local processing space size with each target welding point as the center, and the space is verified. It is verified whether the initial local processing space contains the predecessor and successor points of the target welding point. If so, the initial local processing space is determined as the local processing space of the target welding point. If the predecessor point is missing, it is determined whether it is the first target welding point in the welding path. If so, the initial local processing space is determined as the local processing space of the target welding point. If not, the predecessor point of the target welding point is located according to the welding path and included in the initial local processing space. At the same time, the initial local processing space size is adaptively updated according to the position of the newly added predecessor point, thereby obtaining the local processing space of the current target welding point. If the successor point is missing, it is determined whether it is the last target welding point in the welding path. If so, the initial local processing space is determined as the local processing space of the target welding point. If not, the successor point of the target welding point is located according to the welding path and included in the initial local processing space. The initial local processing space size is also updated synchronously to generate a local processing space that meets the requirements.

[0081] When the TIG welding equipment executes a welding path and moves to the next target welding point, the local processing space update mechanism is triggered. The predecessor point that has been welded is removed from the space, and the original successor point is marked as the current welding point and included in the new successor point to be processed in the welding path.

[0082] The local feature extraction module is used to capture local images of target welds within the local processing space during local processing space updates. It then extracts multimodal features within the local processing space, including the quality features of the predecessor weld, the melt pool features and geometry of the current target weld, and the geometry of the subsequent weld. An industrial camera is used to capture the melt pool image and weld topography of the current target weld at high frequency. Image recognition algorithms are then used to extract quality features such as weld width, penetration uniformity, and porosity defects from the predecessor weld. The melt pool contour, back drag angle, and metal flow state of the current target weld are analyzed. Geometric features such as the surface curvature and diameter variation of the serpentine pipe at the subsequent weld are also acquired.

[0083] Preferably, the specific steps of extracting multimodal features in the local processing space include:

[0084] The welding quality of the precursor point directly reflects the effectiveness of the current welding parameters. By analyzing the penetration state of the precursor point, the current weld parameters can be adjusted in real time. When the local processing space is updated, the precursor point in the current local processing space is located according to the spatial topological relationship between the target weld points. The welding window is set according to the serpentine pipe diameter and welding process requirements. With the precursor point as the center, the welded image within the precursor point welding window is obtained to ensure that the complete weld and heat-affected zone are covered.

[0085] Preprocess the welded image, including grayscale processing, denoising and equalization processing. Use the threshold segmentation algorithm to separate the weld area and the base material area to provide clean data for feature extraction.

[0086] The weld geometry is directly related to the penetration state, and the edge gradient reflects the fusion quality. The weld contour is extracted from the weld area to calculate the weld width. The weld area is skeletonized, and the cross-sectional profile perpendicular to the centerline is calculated after extracting the weld centerline. The weld reinforcement is fitted using the least squares method. The weld edge grayscale gradient amplitude is calculated based on the weld contour to obtain the contour gradient matrix. Morphological operations are used to identify the length of the weld contour void. The weld width is positively correlated with the penetration depth and is used to assess whether the penetration depth meets the standard. Excessive weld reinforcement indicates insufficient penetration, while too low may indicate excessive penetration, reflecting the distribution of the deposited metal. The grayscale gradient amplitude represents the bonding strength between the weld and the base material and is used to detect poor fusion defects. The void length directly indicates the severity of the lack of fusion defect and triggers the quality warning mechanism.

[0087] The geometric features of the area to be welded affect the heat conduction characteristics, and pre-compensation of welding parameters is required. When the local processing space is updated, the target welding point and the subsequent point in the current local processing space are located at the same time. With the target welding point and the subsequent point as the center, the image of the target welding point and the subsequent point in the welding window is obtained;

[0088] After preprocessing the welding image, the geometric features of the target welding point and subsequent points are extracted respectively, including: curvature features, gap width features, and pipe diameter features; the curvature features reflect the degree of bending of the workpiece, affect the heat dissipation rate, and are used to adjust the heat input intensity; the gap width determines the stability and filling requirements of the molten pool, and is used to optimize the wire feeding speed and welding trajectory; the pipe diameter affects the heat capacity distribution and is used to dynamically adjust the current to maintain the consistency of the penetration depth.

[0089] Geometric mutations of adjacent welds can cause fluctuations in penetration depth, requiring early response. The geometric features of the target weld and subsequent welds are compared, and the geometric feature change rate of the subsequent weld relative to the target weld is extracted. The geometric change features of the subsequent welds are constructed, including the curvature change rate, gap width change rate, and tube diameter change rate.

[0090] The molten pool morphology is a direct representation of the penetration depth, and the process stability needs to be monitored in real time. When processing the target weld point in the local processing space, the target weld point is used as the center, and the synchronous trigger circuit is used to ensure that the image acquisition is aligned with the welding arc pulse phase to reduce arc interference and dynamically capture the molten pool image within the welding window of the target weld point.

[0091] The melt pool image is adaptively segmented by threshold and morphological opening operation is combined to remove spatter noise, extract the melt pool contour, calculate the melt pool length, melt pool width and melt pool area, and calculate the melt pool area change rate of the melt pool contour in the melt pool images of adjacent frames; the melt pool aspect ratio characterizes the melt pool morphological stability and is used to judge whether the melt depth is sufficient; the melt pool area change rate reflects the energy input fluctuation and triggers the parameter fine-tuning mechanism to maintain the stability of the melt depth.

[0092] The penetration decision module is used to adaptively update the target penetration according to the multimodal characteristics in the local processing space. By receiving and parsing characteristic data such as the quality of the predecessor point, the current molten pool state and the geometry of the successor point, the current penetration effect is evaluated by comparing with the welding standard. The penetration state is judged in real time based on the molten pool morphology, the welding parameters are pre-compensated based on the geometric characteristics of the successor point, and the control algorithm is used to generate an adjustment strategy. The parameters are output to the welding equipment and the optimization strategy is fed back according to the actual penetration depth. At the same time, emergency treatment is carried out for abnormal penetration depth.

[0093] See also Figure 4 Preferably, the specific steps of adaptively updating the target penetration depth include:

[0094] The TIG welding equipment performs welding processing on the target welding points in the working space according to the welding path, and performs welding operations on the target welding points in the local processing space according to the target penetration value and the initial processing parameters;

[0095] Changes in geometric features such as workpiece curvature and pipe diameter can affect the heat dissipation rate. Preemptive identification of geometric mutation areas can avoid inconsistent penetration due to differences in heat conduction, improving the predictability of welding quality. The multimodal features of the target welding point in the local processing space are acquired. Based on the geometric change features of the subsequent points of the target welding point in the current local processing space, a preset geometric feature change threshold is used to determine whether to perform pre-compensation processing on the subsequent points. If the geometric change features are greater than the preset geometric feature change threshold, then pre-compensation processing is determined for the subsequent points, i.e.:

[0096] Initial compensation coefficients are set for each geometric variation feature, including those for the curvature change rate, gap width change rate, and pipe diameter change rate. Based on the ratio of the geometric variation feature to the corresponding geometric variation threshold, weighted corrections are applied to the initial compensation coefficients for each geometric variation feature. These coefficients serve as compensation coefficients for the geometric variation feature, quantifying their impact on penetration. The initial compensation coefficients provide a baseline adjustment for the geometric feature's impact on penetration. These coefficients are empirically determined based on statistical analysis of the correlation between geometric features and penetration using historical welding data. If pre-compensation is not performed on subsequent points, the target penetration value of the target weld point serves as the initial target penetration value for subsequent points. Quantifying the impact of geometric features on penetration using historical welding data allows for the generation of compensation coefficients that match the degree of geometric variation, avoiding over- or under-compensation.

[0097] The compensation coefficients of each geometric variation feature are weighted and fused to calculate a comprehensive compensation coefficient. This resolves the conflicting compensation strategies under the influence of multiple geometric features. The target penetration value of the current target weld point is multiplied by this comprehensive compensation coefficient to obtain the initial target penetration value of the subsequent weld point, completing the adaptive update of the pre-compensation processing for the target penetration of the subsequent weld point. The weighted fusion weights are set by ranking the features based on their importance, as determined by training with historical welding data.

[0098] After completing the welding operation for the target weld point in a local processing space, the local processing space is automatically updated. The target weld point of the previous local processing space is used as the predecessor point of the current local processing space. The initial target penetration value of the target weld point in the current local processing space is set based on the initial target penetration value of the successor point of the previous local processing space. A time sequence association is established between adjacent weld points, and the current parameters are optimized using historical welding data to improve the consistency of penetration control in long-path welding.

[0099] Relying solely on pre-compensation cannot cover all process fluctuations. Feedback correction through the quality features of the predecessor point is required to avoid the recurrence of defects. The multimodal features of the welded image within the welding window of the predecessor point in the current local processing space are obtained. The quality feature parameters of the predecessor point, including weld width, weld height, contour gradient matrix, and weld contour void length, are extracted. The welding quality status is judged by the quality assessment model, including qualified welding, excessive welding, and missing welding. The confidence score of the welding quality status is output. The contribution of each quality feature parameter to the classification result is analyzed through the SHAP value to generate a quality feature deviation vector. Through the feedback correction of the quality features of the predecessor point, process fluctuations not considered in pre-compensation are covered, reducing the recurrence of welding defects.

[0100] The quality assessment model is used to realize intelligent classification and identification of welding quality status based on the nonlinear mapping relationship of weld quality characteristic parameters, and quantify the influence weight of each feature on the quality status. Through the weld samples under different welding parameters in the historical welding data, including qualified welding samples, excessive welding samples such as excessive penetration and undercut, and missing welding samples such as incomplete fusion and insufficient penetration; the weld width, weld height, contour gradient matrix and weld contour void length of each sample are used as input, and the welding quality status is used as the output label. The quality assessment model is trained through the random forest separation algorithm.

[0101] If the quality assessment model determines that the welding is qualified, the initial target penetration value of the current target welding point is maintained as the target penetration value of the welding process;

[0102] If the quality assessment model determines excessive or missing welds, the dominant quality feature is determined based on the SHAP value. Based on the direction and magnitude of the dominant quality feature deviation vector, a configured nonlinear penetration mapping function is called to calculate the penetration correction coefficient. This correction is then applied to the initial target penetration value of the target weld point, resulting in the target penetration value for the weld process. The nonlinear penetration mapping function maps the dominant quality feature deviation vector into a penetration correction coefficient. This nonlinear transformation quantifies the impact of the feature deviation on penetration through a nonlinear transformation, enabling adaptive correction of target penetration in excessive or missing welds. Historical welding data is collected, using the quality feature deviation vector as input and the calibrated penetration correction coefficient as the output label. The model is trained using nonlinear algorithms such as gradient boosting trees and neural networks, and model hyperparameters are optimized through cross-validation and grid search. Furthermore, by incorporating dynamic factors such as real-time temperature and material thermophysical properties during the welding process, a complete system is constructed, encompassing feature engineering, model training, parameter optimization, and dynamic compensation. This results in a nonlinear mapping function that automatically calculates the penetration correction coefficient based on the deviation vector. This allows for real-time adjustment of penetration for excessive or missing welds, shifting defect repair from offline detection to online correction, shortening the quality feedback cycle.

[0103] Based on the target penetration value of the target welding point in the local processing space, the initial processing parameters are updated through the model of the association between penetration and processing parameters to perform welding processing on the target welding point, and the subsequent points are pre-compensated according to the geometric change characteristics of the subsequent points of the target welding point to obtain the initial target penetration value of the subsequent points until the welding path is completed.

[0104] The risk warning module is used to monitor the fluctuation range of processing parameters such as welding current and arc voltage in real time when the welding process is driven by the target penetration value, and to simultaneously issue a warning of processing quality anomalies based on the multimodal characteristics of the molten pool image; and to issue an intervention anomaly warning based on the welding quality status of the predecessor point in the local processing space and the target penetration correction amount of the subsequent point. When the quality defect of the predecessor point is marked as recurring or the correction amount of the subsequent point exceeds the dynamic threshold, the intervention anomaly warning is activated, indicating a systematic deviation in the process parameters or an abnormal equipment status. At the same time, the parameter initialization module is linked to trigger preventive adjustments.

[0105] Preferably, the specific steps of performing an early warning of abnormal processing quality include:

[0106] Based on the local processing space update mechanism, the melt pool features of the target weld pool image are acquired. Processing and environmental parameters are synchronously collected using an industrial camera and sensor array. Using the spatial coordinate system constructed by the path planning module, the melt pool features are dynamically associated with the target weld location. These features include melt pool length, melt pool width, and melt pool area change rate, enabling data temporal and spatial consistency calibration. The acquisition frequency is linked to the local processing space update cycle to ensure that each image frame corresponds to a unique weld point coordinate.

[0107] The parameter initialization module retrieves the nominal machining parameters corresponding to the target penetration value and calculates the deviation between the real-time machining parameters and the nominal machining parameters. A machining parameter abnormality flag is triggered when the deviation exceeds a preset machining deviation threshold. The machining deviation threshold is set by analyzing the correlation between penetration fluctuations and parameter deviations in historical data.

[0108] The target welding point type and curvature parameters are obtained from the path planning module, the material attributes and pipe diameter dimensions of the parameter initialization module are called, and the molten pool characteristic threshold of the corresponding type of welding point is retrieved from the welding process parameter database. The molten pool characteristic threshold includes the molten pool length range, the molten pool width range, and the molten pool area change rate range;

[0109] For straight welds, the standard ranges of melt pool length and width are configured based on the thermal conductivity characteristics of the material. The basic threshold for the melt pool area change rate is stored in the welding process parameter database. For curved welds, in addition to the material and pipe diameter, the real-time curvature parameters output by the path planning module are synchronously obtained. The threshold ranges of melt pool length and width are adjusted using a pre-configured threshold relaxation strategy. The relaxation range is defined by the material thermal conductivity configuration file, which supports classification management by material type.

[0110] Compare the molten pool characteristics of the target weld point with the corresponding molten pool characteristic threshold and mark the molten pool abnormality. For example, if the molten pool length is greater than the molten pool length range, the molten pool length abnormality flag is triggered. If the molten pool width is greater than the molten pool width range, the molten pool width abnormality flag is triggered. If the molten pool area change rate is greater than the molten pool area change rate range, the molten pool area change rate abnormality flag is triggered.

[0111] According to the abnormal markings of processing parameters and molten pool, the abnormal target welding points are located, and the processing quality abnormality warning is issued according to the position distribution of the abnormal target welding points. The number of abnormal target welding points in the workspace is counted. If it is greater than the preset abnormal point threshold, a processing quality abnormality warning is issued, and the TIG welding equipment is stopped from executing the welding path to weld the target welding points. Otherwise, no processing is performed.

[0112] See also Figure 5 Preferably, the specific steps of intervening in abnormal warning include:

[0113] The welding quality status of the current local processing space precursor point is obtained from the penetration decision module. If it is determined that the welding is excessive or missing, a sliding window is established based on the temporal association of the local processing space to perform defect repeatability analysis. If the number of times the same welding quality status appears in the sliding window is greater than the preset defect threshold, and the confidence level of a single defect exceeds the configured reliability threshold, it is marked as a recurring quality defect.

[0114] The target penetration correction value of the subsequent point output by the penetration decision module is analyzed. If it is greater than the preset correction threshold, it is marked as a correction value exceeding the threshold; and correction detection is performed. The correction monitoring interval is configured, which can be set according to the number of consecutive welds or by time, and is used to count the number of times the correction value exceeds the threshold within the interval; the number of times the correction value exceeds the threshold within the correction monitoring interval is counted. If it is greater than the correction number threshold, an abnormal intervention warning is issued, and the TIG welding equipment is stopped from executing the welding path to weld the target welding point.

[0115] And when the mark of repeated appearance of quality defects of the predecessor point in the local processing space and the mark of the correction amount of the subsequent point exceeding the threshold are triggered at the same time, an intervention abnormality warning is issued, and the TIG welding equipment is stopped from executing the welding path to weld the target welding point.

[0116] Working principle and its effect:

[0117] The present invention constructs a serpentine tube spatial coordinate system and extracts edge features to accurately identify the distribution and curvature parameters of straight and curved segments. Based on the welding point template, it traverses the image sub-area to mark the target welding points, constructs a directed graph with spatial distance, equipment posture change and welding accessibility as weights, and generates a smooth welding path through an intelligent optimization algorithm, thereby reducing the frequency of equipment posture adjustment and collision risk. In the dynamic management of local processing space, the spatial dimensions are differentially defined according to the welding point type combined with parameters such as pipe diameter and curvature. The space is updated in real time by verifying the inclusiveness of the predecessor point and the successor point to ensure the integrity of the multimodal feature extraction. The welded image of the predecessor point is preprocessed to separate the weld and base material areas, and the quality features such as weld width, residual height, and contour gradient matrix are extracted. The geometric feature change rate of the image of the successor point to be welded is compared, and the synchronously collected molten pool length, width, and area change rate are combined to form an adaptive update mechanism for the penetration depth that combines geometric pre-compensation with quality feedback: the compensation coefficient is weighted and corrected based on the ratio of the geometric change feature of the successor point to the threshold, and the comprehensive compensation coefficient is fused to update the initial penetration depth. The dominant features are then analyzed through the predecessor point quality assessment model and the nonlinear mapping function is called to correct the current penetration depth, forming a closed-loop control of prediction, execution, and feedback.

[0118] During welding, the system correlates the melt pool characteristics with the weld point coordinates in real time, comparing the deviation from the rated processing parameters with the melt pool characteristic threshold to trigger anomaly flags. By statistically analyzing the spatial distribution density and cumulative number of anomaly points, a graded early warning system is implemented. If the number of similar defects within the sliding window exceeds the threshold and the confidence level meets the standard, or if the subsequent point correction exceeds the threshold multiple times within the monitoring interval, the system immediately initiates an intervention mechanism to stop welding and activates the parameter initialization module to reset the process parameters. This mechanism effectively addresses the inconsistent penetration depth caused by complex geometric features and uneven heat conduction in serpentine tube welding, significantly improving welding quality stability and production efficiency, reducing welding defects and rework frequency, lowering equipment wear and production costs, and achieving intelligent and precise control of complex structure welding.

[0119] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A serpentine tube TIG welding penetration adaptive control system, characterized in that: include: Initialize the initial processing parameters of the TIG welding equipment according to the basic parameters of the serpentine tube and determine the target penetration value; Acquire a panoramic image of the working area of ​​the TIG welding equipment, identify the straight and curved segments of the serpentine tube, and mark target welding points by traversing image sub-areas of the panoramic image based on a welding point template. Construct a weighted directed graph with the target welding points as nodes, perform path search, and generate a welding path. Classifying the target weld points according to their positions on the serpentine tube, differentially determining the size of the local processing space, defining the predecessor and successor points of the target weld points based on the welding path, verifying and updating the size of the local processing space, and obtaining the local processing space of the target weld points; When the local processing space is updated, the predecessor point, target welding point and successor point in the local processing space are located, images in the welding window are collected respectively, multimodal features in the local processing space are extracted, and the target penetration is adaptively updated: Based on the geometric change characteristics of the subsequent point, the initial compensation coefficient is corrected to update its target penetration depth. The target penetration depth of the initial target welding point is updated according to the local processing space. Then, the welding status is judged based on the quality characteristic parameters of the predecessor point and the current target penetration depth is corrected. When the welding process is driven by the target penetration value, processing quality abnormality warning and intervention abnormality warning are carried out based on the multi-modal characteristics, processing parameters and environmental parameters collected in real time.

2. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of adaptively updating the target penetration depth include: The multimodal features in the local processing space of the target welding point are sequentially obtained, and according to the geometric change features of the subsequent points of the target welding point, a preset geometric feature change threshold is used to determine whether to perform pre-compensation processing on the subsequent points; If so, the initial compensation coefficients of the geometric change features are weightedly modified according to the ratio of the geometric change feature to the corresponding geometric change threshold value, and used as the compensation coefficients of the geometric change features; The compensation coefficients of each geometric change feature are weighted and integrated to calculate the comprehensive compensation coefficient. The target penetration value of the current target welding point is multiplied by the comprehensive compensation coefficient to obtain the initial target penetration value of the subsequent point. If not, the target penetration value of the target welding point is used as the initial target penetration value of the subsequent point.

3. The adaptive control system for serpentine tube TIG welding penetration according to claim 2, characterized in that: The specific step of adaptively updating the target penetration depth further includes: After the welding operation of the target welding point in the local processing space is completed, the local processing space is updated, and the initial target penetration value of the target welding point in the current local processing space is set according to the initial target penetration value of the successor point in the previous local processing space; Extract the quality characteristic parameters of the precursor points, including weld width, weld height, contour gradient matrix, and weld contour void length; use the quality assessment model to determine the welding quality status, including qualified welding, excessive welding, and missing welding, and analyze the contribution of each quality characteristic parameter to the classification result through the SHAP value; The quality assessment model is used to classify and identify the welding quality status based on the nonlinear mapping relationship of the weld quality characteristic parameters, and is obtained through training and fitting based on historical welding data; If the quality assessment model determines that the welding is qualified, the initial target penetration value of the current target weld point is used as the target penetration value of the welding process; if the quality assessment model determines that the welding is excessive or missing, the dominant quality feature is determined according to the SHAP value, and the penetration nonlinear mapping function is called to calculate the penetration correction coefficient according to the direction and size of the dominant quality feature deviation vector, and the initial target penetration value of the target weld point is corrected to serve as the target penetration value of the welding process; the penetration nonlinear mapping function is used to map the deviation vector of the dominant quality feature to the penetration correction coefficient, and is obtained through training and fitting based on historical welding data.

4. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of obtaining the local processing space of the target welding point include: Generate an ordered sequence of target welding points based on the welding path, establish spatial topological associations between the target welding points, and define predecessor points and successor points for each target welding point according to the direction of the welding path; The size of the local processing space is determined based on the type of welding point, namely: The initial local processing space size of the straight welding point is determined by the length of the straight section of the serpentine tube where it is located and the diameter of the serpentine tube; The initial local processing space size of the curved welding point is determined by the length of the curved section of the serpentine tube, the diameter of the serpentine tube and the local curvature characteristics. Taking the target welding point as the center, the initial local processing space is delineated according to the size of the initial local processing space, and the initial local processing space is verified to verify whether the initial local processing space contains the predecessor point and the successor point of the target welding point; If it is included, the initial local processing space is determined as the local processing space; If the predecessor point is missing, the predecessor point of the target welding point is located according to the welding path and is included in the initial local processing space. If the successor point is missing, the successor point of the target welding point is located according to the welding path and is included in the initial local processing space. The initial local processing space size is updated to obtain the local processing space of the target welding point.

5. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of generating the welding path include: Capture panoramic images of the serpentine tube within the workspace of the TIG welding equipment, extract edge features, and construct the serpentine tube spatial coordinate system; Perform straight line detection on edge features, identify the straight and curved segments of the serpentine tube, and calculate the curvature parameters of the curved segments, including the curvature radius and curvature change rate of the curved segments. The curvature mutation points are marked by the preset curvature mutation threshold. Based on the welding point template image, the image sub-region is traversed through a sliding window in the panoramic image, and the normalized cross-correlation coefficient is used to calculate the similarity between the welding point template and the image sub-region. If the similarity is greater than the preset similarity threshold, it is marked as a target welding point.

6. The adaptive control system for serpentine tube TIG welding penetration according to claim 5, characterized in that: The specific steps of generating the welding path also include: According to the spatial coordinate system of the serpentine tube, the coordinate position of the target welding point is located, and according to the position distribution of the target welding point in the straight segment and curved segment of the serpentine tube, the target welding point is classified into a straight welding point or a curved welding point; Based on the classification results of welding points, a weighted directed graph with target welding points as nodes is constructed. The edge weights are determined by the Euclidean distance between target welding points, the change in the posture angle of the TIG welding equipment, and the welding accessibility. A path search is performed on the weighted directed graph with the optimization goal of minimizing the total weight of the path. The initial welding path is generated through selection, crossover and mutation operations, and the welding path of the TIG welding equipment is obtained through smoothing.

7. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of extracting multimodal features in the local processing space include: When the local processing space is updated, the predecessor point in the current local processing space is located, the welding window is set, and the welded image in the predecessor point welding window is obtained with the predecessor point as the center; Pre-process the welded image to separate the weld area from the base material area; For the weld area, the weld contour is extracted to calculate the weld width, and the weld centerline is extracted to calculate the cross-sectional contour perpendicular to the centerline, and the weld residual height is fitted; the grayscale gradient amplitude of the weld edge is calculated according to the weld contour to obtain the contour gradient matrix; the length of the weld contour cavity is identified through morphological operations.

8. The adaptive control system for serpentine tube TIG welding penetration according to claim 7, characterized in that: The specific step of extracting the multimodal features in the local processing space also includes: Locate the target welding point and the subsequent point in the current local processing space, take the target welding point and the subsequent point as the center, and obtain the image to be welded in the welding window of the target welding point and the subsequent point; After preprocessing the welding image, the geometric features of the target welding point and the subsequent points are extracted respectively. The geometric features of the target welding point and the subsequent points are compared, and the geometric feature change rate of the subsequent point relative to the target welding point is extracted to construct the geometric change feature of the subsequent point. When processing the target welding point in the local processing space, the molten pool image in the welding window of the target welding point is dynamically captured through the synchronous trigger circuit with the target welding point as the center; The melt pool image is segmented, the melt pool contour is extracted, the melt pool length, melt pool width and melt pool area are calculated, and the melt pool area change rate of the melt pool contour in the melt pool images of adjacent frames is calculated.

9. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of the processing quality abnormality warning include: Obtaining melt pool features of a melt pool image of a target weld point, the melt pool features including melt pool length, melt pool width, and melt pool area change rate; and associating the melt pool features with the position of the target weld point; Retrieve the rated processing parameters corresponding to the target penetration value, calculate the deviation between the real-time processing parameters and the rated processing parameters, and trigger a processing parameter abnormality flag if the deviation is greater than the preset processing deviation threshold; Set the melt pool characteristic threshold, compare the melt pool characteristics of the target welding point with the corresponding melt pool characteristic threshold, and mark the melt pool abnormality; According to the abnormal processing parameter marks and the abnormal molten pool marks, the abnormal target welding points are located, and the processing quality abnormality warning is issued according to the position distribution of the abnormal target welding points. The number of abnormal target welding points in the workspace is counted. If it is greater than the preset abnormal point threshold, a processing quality abnormality warning is issued.

10. The adaptive control system for serpentine tube TIG welding penetration according to claim 1, characterized in that: The specific steps of intervening in abnormal warning include: The welding quality status of the predecessor point in the current local processing space is obtained. If it is determined that there is excessive welding or missing welding, a sliding window is established based on the temporal correlation of the local processing space to perform defect repeatability analysis. If the number of times the same welding quality status appears in the sliding window exceeds the preset defect threshold, and the confidence level of a single defect is greater than the configured reliability threshold, it is marked as a recurring quality defect. Analyze the target melt depth correction value of the subsequent point. If it is greater than the preset correction threshold, it is marked as the correction value exceeding the threshold; and perform correction detection, configure the correction monitoring interval, and count the number of times the correction value exceeds the threshold within the correction monitoring interval. If it is greater than the correction number threshold, an intervention abnormality warning is issued; and when the repeated appearance mark of the predecessor point quality defect in the local processing space and the subsequent point correction value exceeding the threshold mark are triggered at the same time, an intervention abnormality warning is issued.

Images