Welding quality monitoring method and platform for electro-tricycle frame

Abstract

The invention discloses a welding quality monitoring method and platform for an electro-tricycle frame, and relates to the technical field of welding quality monitoring. The method comprises the steps that three-dimensional scanning is conducted on a frame component of an electro-tricycle, and the welding requirement is determined; welding parameter analysis is conducted, and a combined welding parameter sequence under the periodic sequence is determined; positioning personalized monitoring characteristics of each welding stage in a welding period, and configuring a self-attention sequence; welding data fragments are determined; and abnormal welding judgment and feedback decision analysis are carried out. The technical problems that in the prior art, the whole-period intelligent precise control and dynamic optimization are difficult to achieve in the electric tricycle frame welding process, the welding quality is uneven, the welding efficiency is low, and the welding process stability is insufficient are solved. The technical effects that intelligent precise control and dynamic optimization of the whole period of the welding process of the electro-tricycle frame are achieved, and the welding quality, the welding efficiency and the welding process stability are improved are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of welding quality monitoring, and in particular to a welding quality monitoring method and platform for an electric tricycle frame. Background Art

[0002] In the field of electric tricycle manufacturing, frame welding is a key production link, and its quality and efficiency directly affect the performance and safety of the whole vehicle. However, traditional frame welding technology mainly relies on manual experience operation, which has many disadvantages. On the one hand, in the early stage of welding, due to the lack of accurate component data collection and analysis methods, it is difficult to comprehensively and accurately determine the welding requirements of complex frame components. The differences in welding parameters and process conditions required by different welding structures (such as connections of different materials, welds of various shapes, etc.) have not been fully considered, resulting in a relatively extensive welding process that cannot meet the needs of refined production. On the other hand, during the welding process, the control of welding parameters is not accurate and stable enough, and it is difficult to make real-time and dynamic adjustments according to the actual welding situation. At the same time, the welding monitoring method is single and lagging, and most of them only rely on manual visual inspection of some appearance features. It is impossible to conduct comprehensive and timely monitoring of key parameters and internal quality in the welding process, and it is difficult to detect potential welding defects in advance, resulting in a high defective rate, which seriously affects the welding quality and production efficiency. In addition, the stability of the welding process is poor, and the welding quality of different batches of products fluctuates greatly, making it difficult to ensure the consistency and reliability of the performance of the whole vehicle, and it is difficult to adapt to the requirements of modern manufacturing industry for high-quality and high-efficiency production.

[0003] The existing technology has technical problems such as difficulty in achieving full-cycle intelligent, precise control and dynamic optimization of the electric tricycle frame welding process, uneven welding quality, low welding efficiency and insufficient welding process stability. Summary of the invention

[0004] The present application provides a welding quality monitoring method and platform for an electric tricycle frame, which is used to solve the technical problems in the prior art that it is difficult to achieve full-cycle intelligent and precise control and dynamic optimization of the welding process of the electric tricycle frame, the welding quality is uneven, the welding efficiency is low, and the welding process stability is insufficient.

[0005] In view of the above problems, the present application provides a welding quality monitoring method and platform for an electric tricycle frame.

[0006] In a first aspect of the present application, a method for monitoring the welding quality of an electric tricycle frame is provided, the method comprising: The frame components of the electric tricycle are scanned in three dimensions to determine the welding requirements under the welding cycle of the frame, wherein different welding structures have different welding requirements; based on the welding requirements, welding parameter analysis is performed in combination with a welding decision model to determine a combined welding parameter sequence under a cycle sequence; personalized monitoring features of each welding stage under a welding cycle are located according to changes in welding requirements, and a self-attention sequence is configured, wherein personalized monitoring features are determined by welding key parameters-expected welding features; the combined welding parameter sequence and the self-attention sequence are mapped, an automatic monitoring module is set, welding control is performed under a welding cycle by controlling the combined welding parameter sequence, and based on the mapping relationship, personalized monitoring features based on the self-attention sequence are triggered, personalized monitoring and data screening are performed, and welding data fragments are determined; based on the welding decision model, abnormal welding judgment and feedback decision analysis are performed on the welding data fragments, and welding control and welding monitoring are reversely guided.

[0007] The second aspect of the present application provides a welding quality monitoring platform for an electric tricycle frame, the platform comprising: A welding requirement determination module is used to perform a three-dimensional scan on the frame components of the electric tricycle to determine the welding requirements under the welding cycle of the frame, wherein different welding structures have different welding requirements; a combined welding parameter sequence determination module is used to perform welding parameter analysis based on the welding requirements and in combination with a welding decision model to determine the combined welding parameter sequence under the cycle sequence; a self-attention sequence configuration module is used to locate the personalized monitoring features of each welding stage under the welding cycle based on the changes in welding requirements and configure the self-attention sequence, wherein the personalized monitoring features are determined by welding key parameters-expected welding features; a welding data fragment determination module is used to map the combined welding parameter sequence with the self-attention sequence, set an automatic monitoring module, perform welding control under the welding cycle by controlling the combined welding parameter sequence, trigger the personalized monitoring features based on the self-attention sequence based on the mapping relationship, perform personalized monitoring and data screening, and determine welding data fragments; an abnormal welding determination module is used to perform abnormal welding determination and feedback decision analysis on the welding data fragments based on the welding decision model, and reversely guide welding control and welding monitoring.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: The welding demand determination module is used to perform three-dimensional scanning on the frame components of the electric tricycle to determine the welding demand under the welding cycle of the whole frame; the combined welding parameter sequence determination module performs welding parameter analysis based on the welding demand and determines the combined welding parameter sequence under the cycle sequence; the self-attention sequence configuration module is used to locate the personalized monitoring features of each welding stage under the welding cycle according to the changes in welding demand and configure the self-attention sequence; the welding data fragment determination module is used to map the combined welding parameter sequence with the self-attention sequence, set the automatic monitoring module, and perform welding control under the welding cycle by controlling the combined welding parameter sequence. Based on the mapping relationship, the personalized monitoring features based on the self-attention sequence are triggered, and personalized monitoring and data screening are performed to determine the welding data fragments; the abnormal welding judgment module performs abnormal welding judgment and feedback decision analysis on the welding data fragments, and reversely guides welding control and welding monitoring. The technical effect of realizing intelligent and precise control and dynamic optimization of the entire cycle of the electric tricycle frame welding process and improving welding quality, welding efficiency and welding process stability is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic flow chart of a welding quality monitoring method for an electric tricycle frame provided in an embodiment of the present application.

[0011] Figure 2 A schematic diagram of the structure of a welding quality monitoring platform for an electric tricycle frame provided in an embodiment of the present application.

[0012] Explanation of reference numerals: welding requirement determination module 10 , combined welding parameter sequence determination module 20 , self-attention sequence configuration module 30 , welding data fragment determination module 40 , abnormal welding determination module 50 . DETAILED DESCRIPTION

[0013] This application provides a welding quality monitoring method and platform for an electric tricycle frame, aiming to solve the technical problems in the prior art of difficulty in achieving full-cycle intelligent, precise control and dynamic optimization of the welding process of an electric tricycle frame, uneven welding quality, low welding efficiency and insufficient welding process stability.

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0015] Embodiment 1, as Figure 1 As shown, the present application provides a method for monitoring the welding quality of an electric tricycle frame, the method comprising: Step S100: Perform a three-dimensional scan on the frame components of the electric tricycle to determine the welding requirements of the frame under the welding cycle of the entire vehicle, wherein different welding structures have different welding requirements.

[0016] Specifically, in the initial stage of the welding quality monitoring of the electric tricycle frame, high-precision three-dimensional scanning equipment is used to carry out comprehensive and detailed scanning operations on the frame components. The electric tricycle frame is composed of many different welded components, such as the main beam, sub-beam, support parts and connectors. These components are distributed in different positions of the frame and bear different mechanical functions. Therefore, their welding structures show significant diversity. For example, the main beam welding structure located in the key load-bearing part of the frame needs to bear a large load force, so the corresponding weld shape is usually designed as a continuous and thick butt weld, and the weld size is required to be wider to ensure sufficient welding strength; while in some non-load-bearing auxiliary connection parts, fillet welds are used, and their size is relatively small. At the same time, the materials of different components are also different. Common ones are ordinary carbon steel, alloy steel, etc., and there are differences in thickness, which further leads to changes in welding requirements. For thinner low-carbon steel components, heat input must be precisely controlled during welding to prevent burn-through. For thick-walled alloy steel components, the effect of alloying elements on welding performance must be considered, and appropriate welding process parameters must be selected to ensure weld quality and achieve corresponding strength, toughness and other welding standards, such as specific weld penetration and width requirements, as well as mechanical properties such as hardness and tensile strength. By accurately determining these complex and diverse welding requirements, a solid foundation can be laid for subsequent welding parameter setting and quality monitoring.

[0017] Step S200: Based on the welding requirement, welding parameter analysis is performed in combination with a welding decision model to determine a combined welding parameter sequence under a periodic sequence.

[0018] Specifically, the basic welding parameter data sets that accurately match the key factors such as the material of the frame components, the shape and position of the weld are retrieved from the database of the welding decision model. These data sets are derived from the deep mining of a large amount of historical welding process data and the optimization processing of machine learning algorithms, covering key information such as the thermophysical performance parameters of various materials under different welding conditions, the critical data of metallographic structure transformation, and the boundary value of the welding process window. For specific welding structures, such as the butt weld of the frame main beam, the reasonable range of heat input is accurately calculated based on the chemical composition of its material (such as carbon equivalent, alloy element content), and the initial value range of welding current and voltage that can ensure that the weld depth and width meet the design requirements is determined through the coupling operation of the heat conduction equation and the molten pool fluid mechanics model. At the same time, considering the accessibility and space limitations of the welding position, the robot kinematic model and the association algorithm of the welding process parameters are used to optimize the welding speed to ensure that the welding gun posture is stable and the welding process is smooth. In terms of shielding gas parameter setting, based on the activity of welding materials and the sensitivity of welds to impurities such as oxidation and nitridation, combined with the principles of gas dynamics and welding pool protection theory, the type, flow rate and nozzle structure parameters of shielding gas are accurately calculated to build a stable and effective gas protection atmosphere to prevent defects such as pores and cracks in welds. Through the refined debugging and optimization combination of the above parameters, based on the intelligent reasoning and iterative calculation mechanism of the welding decision model, the combined welding parameter sequence under the complete cycle sequence is finally determined to ensure that a set of welding parameters corresponding to each sequence node can accurately meet the complex process requirements of the corresponding welding structure, providing solid technical support for the precise control and high-quality implementation of the subsequent welding process.

[0019] Step S300: locate the personalized monitoring features of each welding stage in the welding cycle according to the change of welding requirements, and configure the self-attention sequence, wherein the personalized monitoring features are determined by the welding key parameters-expected welding features.

[0020] Specifically, as the welding process gradually advances from the initial stage, the welding requirements of different welding components and parts will change significantly due to factors such as material properties, structural design and process requirements. For example, when splicing the basic frame of the frame at the beginning of welding, due to the initial connection of multiple components, the welding requirements focus on ensuring the positioning accuracy and initial fusion quality between the components. At this time, the key welding parameters such as the stability of the welding current, the uniformity of the welding speed and the starting angle of the welding gun become the focus of attention. Through real-time monitoring and precise control of these key parameters, the expected welding characteristics are seamless docking between the components and uniform forming of the initial weld, and then these key welding parameters are closely associated with the expected welding characteristics to determine the personalized monitoring characteristics of this stage. When welding enters the key reinforcement welding stage, such as the reinforcement welding of the frame load-bearing beam, due to the extremely high requirements for welding strength and internal quality of the weld, the dynamic matching of welding current and voltage and the precise control of welding heat input become the core key parameters. The expected welding characteristics are transformed into high-strength, low-defect high-quality welds, including the control of weld penetration, weld width, internal pores and cracks. Through the close monitoring of these key parameters, the welding process is adjusted with real-time feedback to ensure that the expected welding characteristics are achieved, thereby determining the personalized monitoring characteristics of this stage. Based on these personalized monitoring features determined at different welding stages, the self-attention mechanism algorithm is used to assign corresponding weights and focus to each feature, and construct a self-attention sequence. This sequence can automatically focus on key monitoring features according to the importance and real-time status of each stage during the welding process, dynamically adjust monitoring resources and analysis focus, thereby realizing refined and intelligent monitoring of the entire welding cycle, providing a strong guarantee for timely discovery and resolution of potential welding quality problems, and ensuring the stability and reliability of the welding quality of the electric tricycle frame.

[0021] Step S400: Map the combined welding parameter sequence and the self-attention sequence, set an automatic monitoring module, perform welding control under the welding cycle by controlling the combined welding parameter sequence, trigger the personalized monitoring features based on the self-attention sequence based on the mapping relationship, perform personalized monitoring and data screening, and determine welding data fragments.

[0022] Specifically, it is necessary to first complete the precise mapping of the combined welding parameter sequence and the self-attention sequence, and closely link the parameter setting and monitoring focus in the welding process. Through a highly intelligent programming algorithm, each set of parameters in the combined welding parameter sequence is associated with the corresponding personalized monitoring features in the self-attention sequence one by one, so that at each stage of the welding process, when the welding operation is carried out according to the predetermined combined welding parameter sequence, the personalized monitoring features based on the self-attention sequence can be synchronously triggered according to this mapping relationship. Immediately, the automatic monitoring module began to operate efficiently. It relies on a high-precision sensor network, such as current sensors, voltage sensors, visual sensors, and temperature sensors, to conduct all-round, multi-dimensional real-time monitoring of the welding process. In terms of welding control, the current, voltage, welding speed and other parameters output by the welding equipment are precisely controlled in strict accordance with the set values ​​of the combined welding parameter sequence to ensure that the welding process strictly follows the predetermined process path. At the same time, based on the triggered personalized monitoring features, the automated monitoring module conducts real-time analysis and screening of the massive amount of welding data collected. This data screening process is equivalent to a sophisticated data preprocessing step. It uses filtering algorithms, feature extraction algorithms, and data classification algorithms to accurately identify the effective data parts that are closely related to the personalized monitoring features from the original data, and removes irrelevant data noise interference caused by environmental noise, slight equipment jitter, and other factors, thereby determining welding data fragments that can truly and accurately reflect the welding quality status. These data fragments will serve as the key basis for welding quality judgment and decision analysis in subsequent steps, providing a solid data foundation for the reliability and accuracy of the entire welding quality monitoring system, ensuring that any subtle quality problems that may occur in the welding process can be discovered in a timely and effective manner, and providing strong support for subsequent optimization and adjustment.

[0023] Step S500: Based on the welding decision model, abnormal welding judgment and feedback decision analysis are performed on the welding data fragments to reversely guide welding control and welding monitoring.

[0024] Specifically, the algorithm based on support vector machine (SVM) is used to perform abnormal welding judgment and feedback decision analysis on welding data fragments to achieve effective control of the welding process. First, a large amount of historical welding data is collected, including welding parameters (such as current, voltage, welding speed, etc.) in normal welding processes and various known abnormal welding conditions, as well as corresponding welding quality assessment results (such as weld strength, defect type and degree, etc.). These data are divided into training sets and test sets, and the data in the training set are used to train the SVM model. For welding data fragments, their key feature vectors are extracted. These feature vectors contain the values ​​of various key parameters in the welding process and their change trend information. These feature vectors are input into the trained SVM model. The SVM model maps the low-dimensional feature vectors to the high-dimensional space through its kernel function (such as the Gaussian kernel function), and finds an optimal classification hyperplane in the high-dimensional space to distinguish normal welding data from abnormal welding data. When the SVM model determines that there is an abnormal welding situation in the welding data fragment, feedback decision analysis is performed based on the pre-set decision rules. For example, if it is determined that the weld penetration is insufficient due to unstable welding current, the current value that needs to be adjusted is calculated based on the relationship model between current and penetration learned in the SVM model, and a suitable adjustment rate and adjustment timing are determined in combination with the characteristics of the current welding stage and the welding structure. These adjustment information is fed back to the welding control system to adjust the welding current parameters in real time. At the same time, the monitoring parameters of the automatic monitoring module are adjusted according to the type and severity of the abnormality. For more serious abnormalities, such as possible crack risks, the frequency and accuracy of monitoring the microstructure of the weld are increased, such as using high-resolution ultrasonic detection or X-ray detection technology for local key monitoring; for general abnormalities, such as minor porosity problems, the monitoring sensitivity of welding current and molten pool fluctuations is appropriately increased to promptly detect whether the problem has been effectively solved, thereby forming a closed-loop feedback control mechanism based on the SVM algorithm to ensure that the welding quality meets the requirements, reversely accurately guide welding control and welding monitoring, and ensure the welding quality and overall performance of the electric tricycle frame.

[0025] In a possible implementation, step S300 further includes: Step S310: for the welding cycle, locate the welding position of the component with the changed welding demand as the switching position of the welding monitoring, and the welding position of the component is the initial position for entering the next welding stage.

[0026] Step S320: Mapping the switching position with the personalized monitoring feature to determine the monitoring switching condition.

[0027] Step S330: Based on the monitoring switching condition, the self-attention sequence is marked.

[0028] Specifically, as the welding work progresses, the welding process will be carried out in different stages, and each stage has unique welding requirements, which cover many aspects such as welding parameters, weld quality standards, and material property adaptation. When welding transitions from one stage to the next, it often means that the welding requirements have changed significantly. For example, when spot welding the simple frame structure of the frame at the beginning, it is only necessary to ensure that the components are initially positioned and fixed, and the welding current, voltage and other parameters are relatively low. The requirements for the appearance and strength of the weld are in the initial stage; but when entering the new stage of continuous welding reinforcement of the key load-bearing parts of the frame, a higher welding current is required to ensure sufficient penetration depth, and at the same time, there are more stringent requirements for the internal quality, strength and appearance flatness of the weld. It is at the key nodes where these welding requirements change significantly, that is, the initial component welding position entering the next welding stage, that it is accurately located and set as the switching position for welding monitoring. The purpose of this setting is that when the welding process reaches these positions, it can trigger the switching action of monitoring, so that the monitoring work can be flexibly adjusted according to the corresponding personalized features in the self-attention sequence, thereby ensuring that in each different welding stage, it can focus on the most critical and most suitable features for the actual welding needs of this stage for accurate monitoring, grasp the welding quality in real time, and escort the high-quality welding of the entire electric tricycle frame.

[0029] Based on the decision tree algorithm, the switching position and personalized monitoring features are mapped to determine the monitoring switching conditions. First, a large amount of historical welding data is collected, which covers various parameter information and corresponding welding quality conditions at different welding stages and different switching positions, and is used as a training set. The decision tree algorithm learns from these data, takes the changes in welding stages and the characteristics of switching positions (such as material thickness changes, welding structure changes, etc.) as decision nodes, and gradually builds a decision tree model. For example, when encountering a switching position from a simple butt weld to a complex corner weld in frame welding, the decision tree model will analyze the material properties and welding angle changes at this position based on the learned knowledge, and associate it with the corresponding personalized monitoring features (such as the dynamic adjustment range of welding current, the precise control value of welding gun angle, the expected geometric characteristics of weld formation, etc.). By continuously recursively dividing the training data, the monitoring switching conditions under different branch conditions are determined. When the actual welding process reaches a specific switching position, the current welding parameters and position information are input into the decision tree model. Based on the established decision rules, the model quickly and accurately outputs the corresponding monitoring switching conditions, such as whether it is necessary to increase the monitoring frequency of the molten pool fluidity, whether to adjust the flow rate of the shielding gas, etc., thereby achieving refined monitoring and control of the welding process and ensuring that the welding quality meets the requirements.

[0030] The self-attention sequence is identified by combining the dynamic time warping (DTW) algorithm with the weighted labeling method. First, for each monitoring switching condition, it is converted into a time series feature vector, which contains key elements such as welding parameter change trend, position information and corresponding timestamp. At the same time, each element in the self-attention sequence is also subjected to feature extraction and time series transformation in the same way. Then, the DTW algorithm is used to calculate the optimal matching path and similarity between the time series of the monitoring switching condition and the time series of each element in the self-attention sequence. The DTW algorithm can find the minimum cumulative distance between the two sequences while allowing a certain time expansion and contraction through the dynamic programming method, thereby determining the most matching self-attention sequence element or element combination. Once the matching self-attention sequence part is found, the corresponding weights and labels are assigned to these matching elements according to the pre-set weighting strategy. For example, if the monitoring switching condition is highly matched with a self-attention sequence element and the condition has a significant impact on the welding quality, it is given a higher weight and a specific identification symbol is added, such as adding a marker bit in the data structure or modifying a specific attribute value, to indicate the importance of the element under the current monitoring switching condition and the degree of focus. In this way, in the subsequent welding monitoring process, it is possible to quickly locate and focus on the key self-attention sequence elements based on these identifiers, realize accurate and dynamic monitoring of the welding process, effectively improve the efficiency and accuracy of welding quality monitoring, and ensure the high-quality completion of the electric tricycle frame welding.

[0031] In a possible implementation, step S400 further includes: Step S410: Determine a first welding parameter combination based on the combined welding parameter sequence.

[0032] Step S420: Perform the first stage of welding control through the first welding parameter combination, determine the first self-attention feature based on the mapping relationship, control the monitoring equipment to perform synchronous welding monitoring, and determine the welding data stream, wherein monitoring is performed through collaborative vision, laser detection and control data stream.

[0033] Step S430: Screening the welding data stream to determine the welding data fragments.

[0034] Specifically, the combined welding parameter sequence is deeply analyzed. This sequence is determined based on the previous welding requirements for the electric tricycle frame and the welding decision model, and contains accurate welding parameter information for each stage of the entire welding cycle. From it, a set of parameters suitable for the initial stage of welding is extracted, namely the first welding parameter combination, which covers key parameters such as the starting value of the welding current, the setting range of the welding voltage, the preliminary value of the welding speed, and the initial flow rate of the shielding gas. These parameters are the basis for ensuring that the welding process starts smoothly and meets the initial welding requirements of the frame.

[0035] According to the precisely determined first welding parameter combination, it is input into the advanced welding control system to start the first stage of welding operation. Parameters such as welding current, voltage and welding speed are accurately output according to the preset values ​​to ensure that the welding gun performs welding operations stably and accurately at the corresponding position of the frame, laying the foundation for the subsequent formation of high-quality welds. At the same time, relying on the mapping relationship between the combined welding parameter sequence and the self-attention sequence constructed in the early stage, the first self-attention feature corresponding to this stage is quickly locked. For example, when the first welding parameter combination is set to low current and slow welding speed to ensure the initial fusion quality of the key connection parts of the frame, the first self-attention feature mapped to it focuses on the slow and uniform formation of the molten pool, the fine positioning of the starting section of the weld, and the initial microscopic performance of the fusion of the welding material and the base material. Subsequently, a variety of monitoring equipment were controlled to carry out synchronous welding monitoring operations. A high-resolution collaborative visual monitoring system was used to capture real-time images of the welding area from multiple angles, accurately identifying whether there were signs of defects such as pores, cracks, and undercuts on the weld surface; with the help of precise laser detection technology, the welding area was penetrated to detect possible internal defects such as unfused and slag inclusions, and key information such as the temperature distribution inside the molten pool and the metal flow state were fed back in real time; at the same time, the electrical data flow in the welding control process was closely combined to monitor and analyze the actual fluctuations of the welding current and voltage and the stability of the wire feeding speed in real time. By integrating and synchronously processing the data obtained by these different monitoring methods, a full-scale, multi-dimensional welding data stream is formed. This data stream covers rich information such as surface forming conditions, internal quality conditions, and equipment operating parameters during the welding process, providing a data basis for the subsequent accurate evaluation of welding quality and abnormal judgment, and effectively ensuring the high quality and reliability of the electric tricycle frame welding.

[0036] According to the preset first self-attention features and welding quality standards, the rules and thresholds for data screening are set, such as eliminating abnormal values ​​of visual monitoring data caused by ambient light interference, filtering out noise data generated by welding spatter in laser detection, and screening out data fragments that are closely related to the first welding parameter combination and can truly reflect the key status of the welding process. Then, the data that has been initially screened is further integrated and refined, redundant information is removed, and data that plays a key role in evaluating welding quality is extracted and sorted, and finally welding data fragments that can accurately reflect the welding quality status of the first stage are determined. These data fragments will serve as the core basis for subsequent abnormal welding judgment and quality analysis, laying a solid data foundation for ensuring the welding quality of the entire frame.

[0037] In a possible implementation, step S500 further includes: Step S510: Identify the welding data fragments, and integrate them to determine the first welding key parameter and the second welding feature.

[0038] Step S520: extracting key welding parameters based on the first welding parameter combination, performing parameter response determination on the first key welding parameters, and determining a first monitoring result.

[0039] Step S530: Based on the second expected welding feature, a welding effect determination is performed on the second welding feature to determine a second monitoring result.

[0040] Step S540: Determine a welding feedback strategy according to the first monitoring result and the second monitoring result.

[0041] Specifically, for the collected welding data fragments, it is necessary to use data preprocessing technology to organize and normalize them so that they can be suitable for subsequent SVM algorithm analysis. Through feature engineering methods, the key information contained in these data fragments can be identified, and various data dimensions that can reflect the core conditions of the welding process can be extracted. For example, from the data sequence of current, voltage, etc. that changes with time, the average current value, current fluctuation amplitude, voltage stability time, etc. are extracted as the first key welding parameters; from the weld image features and the molten pool related data reflected by laser detection, the weld width, uniformity of penetration, number of surface defects, etc. are summarized as the second welding features. After these parameters and features are quantified and organized, they form the input feature vector required by the SVM algorithm, laying the foundation for subsequent judgment work.

[0042] Based on the first welding parameter combination set previously, this is equivalent to a set of standard reference values ​​under ideal conditions. The numerous key welding parameters extracted from the actual welding process constitute the sample data to be analyzed. The SVM algorithm is used to construct a binary classification model (such as normal response and abnormal response classification), and the various parameter values ​​in the first welding parameter combination and the corresponding tolerance range are used as positive sample features, and the actual extracted key welding parameters are input into the SVM model as test sample features. SVM uses its core hyperplane division principle and based on the trained model (the training process uses a large amount of marked historical welding parameter data, marked as normal response or abnormal response) to determine the degree of deviation of the actual key welding parameters from the first welding parameter combination, thereby determining whether the first monitoring result is a normal parameter response or an abnormal situation. If the test sample falls within the normal response area divided by the hyperplane, the first monitoring result is judged to be normal; otherwise, it is judged to be abnormal.

[0043] The second expected welding feature represents the ideal welding effect state that meets the requirements of high-quality welding. SVM can also be used to build a classification model to determine the welding effect. The ideal sample data corresponding to the second expected welding feature that meets the standard (such as the ideal weld width, penetration depth and other data combinations) is used as the positive sample, and the second welding feature data obtained and sorted during the actual welding process is used as the test sample to be classified. The SVM model classifies and judges the actual second welding feature based on the difference characteristics between samples of different categories learned during pre-training (obtained by marking and learning a large number of historical samples containing different welding effects), to see whether it is closer to the ideal welding effect or there is a deviation, and then determine whether the second monitoring result is a good or poor welding effect, so as to evaluate the degree of compliance between the actual welding results and expectations.

[0044] According to the first monitoring result and the second monitoring result obtained above, a welding feedback strategy is formulated by comprehensive consideration. Since the SVM algorithm can not only give a category judgment in the classification process, but also output relevant information such as the distance of the sample from the classification hyperplane (which can be understood as a manifestation of confidence or degree of deviation). Based on this information, the degree of abnormal parameter response in the first monitoring result and the degree of poor welding effect in the second monitoring result can be quantitatively analyzed. For example, if the first monitoring result shows that the parameter response deviates seriously from the normal range (far away from the positive class hyperplane), and the second monitoring result shows that the welding effect is also poor (also far away from the hyperplane corresponding to the ideal welding effect), then the feedback strategy may be to adjust the welding parameters to a large extent, such as changing the set values ​​of current and voltage, and at the same time improve the welding process, such as increasing the number of welding layers or optimizing the welding sequence; if the degree of deviation is small, relatively mild feedback strategies such as fine-tuning welding parameters and strengthening local monitoring can be adopted. Through this quantitative analysis based on the judgment results of the SVM algorithm, the appropriate welding feedback strategy can be determined more accurately and scientifically to achieve effective regulation and optimization of the welding quality of the electric tricycle frame.

[0045] In a possible implementation, step S540 further includes: Step S541: If there is an abnormality in the first monitoring result or the second monitoring result, a first feedback decision is made to determine a first feedback strategy.

[0046] Step S542: proofread and verify the first monitoring result and the second monitoring result, make a second feedback decision, and determine a second feedback strategy, wherein whether the first monitoring result and the second monitoring result are consistent is used as the proofreading and verification method, and the disturbance factor of the welding process under inconsistent proofreading and verification results is used as the feedback decision target.

[0047] Step S543: integrating the first feedback strategy and the second feedback strategy to determine the welding feedback strategy.

[0048] Specifically, the abnormality inspection mechanism is first activated. Once an abnormality is found in the first monitoring result (i.e., the response of welding parameters, such as the deviation between the set control current and the actual response current exceeds the allowable range, or the voltage stability does not meet the expected standard, etc.) or the second monitoring result (such as the appearance defects such as pores and cracks in the weld, or the welding effect indicators such as the penetration depth and width do not meet the design requirements), the first feedback decision is immediately made according to the preset rules and empirical models. This involves direct adjustment of welding parameters, such as calculating the current value, voltage correction amount or welding speed fine-tuning value to be compensated according to the direction and size of the parameter deviation, and considering the material properties and process requirements of the current welding stage, determining whether to make adjustments immediately or wait for a specific welding node to make adjustments, thereby determining the initial first feedback strategy, which aims to quickly correct obvious abnormal parameters or effect problems and get the welding process back on track.

[0049] Data is obtained from multiple different types of sensors, including Hall sensors for monitoring welding current and voltage, encoders for measuring welding speed, visual cameras for detecting the appearance of welds, and laser sensors for detecting the internal conditions of the molten pool. Through sensor data fusion technology, these data from different data sources and different physical quantities are preprocessed to synchronize them in time and space, unify the data format and coordinate system, and form a complete data description of the welding process. Then, the fused data is deeply analyzed using the association rule mining algorithm and cluster analysis algorithm in the machine learning algorithm. Association rule mining is used to discover the potential association pattern between the welding parameters (such as current, voltage, welding speed, etc.) in the first monitoring result and the welding effect characteristics (such as weld width, penetration, surface defects, etc.) in the second monitoring result. For example, if it is found that when the welding current fluctuates within a specific range, the change in weld penetration shows a certain specific law, then this association relationship will be extracted. Cluster analysis divides data points with similar characteristics into different clusters to more clearly identify abnormal situations and potential disturbance factors. When the first monitoring result is found to be inconsistent with the second monitoring result, for example, through cluster analysis, it is found that the welding parameters in a set of data are within the normal range, but the penetration depth and appearance quality of the weld are obviously deviated from the expected cluster, which indicates that there are disturbance factors. At this time, the fault tree analysis (FTA) technology is used to construct a fault tree model from multiple aspects such as welding materials, welding equipment, welding environment and operating process, and the factors that may cause this inconsistency are checked layer by layer. For example, consider whether the chemical composition of the welding material is uniform, whether the impurity content exceeds the standard; the power supply stability of the welding equipment, the wear of the electrode; the temperature, humidity, and electromagnetic field interference of the welding environment; and whether the welding angle and welding sequence in the operating process meet the specifications. Based on the disturbance factors determined by the fault tree analysis, make targeted second feedback decisions. If it is a welding material problem, take measures such as changing the material supplier and pre-treating the material; for welding equipment problems, such as unstable power supply, calibrate the power supply or replace the power module; if it is an environmental factor, install temperature and humidity adjustment equipment, electromagnetic shielding devices, etc.; for operating process problems, train and guide operators to optimize welding process parameters and operating procedures. These measures are combined to form a second feedback strategy, which ensures stable and reliable welding quality by eliminating or reducing disturbance factors in the welding process, making the first monitoring results consistent with the second monitoring results, effectively improving the quality and stability of electric tricycle frame welding, and ensuring the efficiency and reliability of the entire welding production process.

[0050] The first feedback strategy and the second feedback strategy determined earlier are organically integrated, and the adjustment measures, implementation timing and impact on the overall welding process of the two strategies are comprehensively considered. Possible duplication or conflicting parts are removed, and the combination is optimized to form a complete set of welding feedback strategies. This strategy can not only effectively correct the abnormal situations that have occurred, but also prevent the recurrence of potential problems. By accurately controlling the welding parameters, optimizing the welding environment and equipment status, it ensures that the welding process can be steadily carried out according to the predetermined quality standards, and realizes the all-round and dynamic control of the welding quality of the electric tricycle frame, effectively improving the overall welding quality and performance of the frame, reducing the scrap rate and quality risks, and ensuring the safety and reliability of the product.

[0051] In a possible implementation, step S543 further includes: Step S5431: The first feedback strategy includes a first compensation strategy based on parameter response deviation and a second compensation strategy based on welding effect deviation, and the second feedback strategy is a compensation strategy based on disturbance factors, wherein the parameter response deviation is the difference between the welding key parameter in the first welding parameter combination and the first welding key parameter.

[0052] Specifically, the first feedback strategy is a preliminary adjustment plan formulated for abnormal conditions that occur during the welding process. It is further subdivided into a first compensation strategy based on parameter response deviation and a second compensation strategy based on welding effect deviation. Among them, the parameter response deviation originates from the difference between the first welding parameter combination (which is the ideal parameter set for the initial welding setting) and the first welding key parameter extracted during the actual welding process. For example, the standard value of the welding current set by the first welding parameter combination is 150A, and the current value of the first welding key parameter actually monitored is 140A on average within a certain period of time, then the difference of 10A is the parameter response deviation. The first compensation strategy formulated based on this deviation involves accurately calculating the amount of current compensation that needs to be increased, and combining the characteristics of the welding process and equipment to determine at what rate and in what way to compensate this part of the current, so that the actual welding parameters can be closer to the preset ideal values, thereby improving the stability and quality of the welding process. At the same time, when the welding effect deviates, such as the penetration of the weld does not meet the expected standard, or appearance defects such as pores and cracks appear, the second compensation strategy formulated based on these welding effect deviations will be activated. This includes additional adjustments to the welding process parameters, such as appropriately reducing the welding speed to increase the molten pool residence time, thereby improving the penetration depth; or adjusting the flow and composition of the shielding gas to reduce the generation of pores. The second feedback strategy focuses on compensating for disturbances that cause abnormalities in the welding process. These disturbances may come from changes in the characteristics of the welding material (such as the deviation between the actual composition of the material and the standard composition affecting the thermal physical properties of the welding), fluctuations in the welding environment (such as environmental factors such as temperature, humidity, and electromagnetic fields that interfere with the stability of the welding process), and changes in the performance of the welding equipment (such as electrode wear, instability of the wire feeding mechanism, etc.). Based on the precise analysis and diagnosis of these disturbance factors, corresponding compensation strategies are formulated. For example, if it is found that the thermal conductivity of the welding material increases due to composition deviation, resulting in excessive heat loss during the welding process, affecting the formation of the molten pool and the quality of the weld, then the second feedback strategy includes recalibrating the welding current and voltage to compensate for the heat lost due to changes in the thermal conductivity of the material, or pretreating the material to restore its thermal conductivity to a normal range, thereby ensuring that the welding process can overcome these potential interference factors, ensuring the stability and reliability of the welding quality, and enabling the entire welding process to be carried out more accurately according to the expected goals, thereby improving the overall quality and performance level of the electric tricycle frame welding.

[0053] In a possible implementation, step S100 further includes: Step S110: determining the scanning data of the vehicle frame component and locating the welding component data by performing three-dimensional scanning, wherein the welding component data includes weld features.

[0054] Step S120: Classifying the welding component data based on preset variables of component structural characteristics to determine the welding component class.

[0055] Step S130: traverse the welding component classes, configure component welding requirements, and perform positive serialization integration based on the welding sequence to determine the welding requirements under the vehicle frame welding cycle.

[0056] Specifically, advanced 3D scanning technology is used to perform all-round, high-precision scanning operations on the frame components of the electric tricycle. During this process, equipment such as laser scanners or structured light scanners will collect data on the surface of the components from multiple angles to accurately determine the scanning data of the frame components. These data cover detailed information such as the geometry, size, and surface flatness of the components. Based on the acquisition of scanning data, the welding component data is accurately located through specialized image processing algorithms and feature recognition technologies, with a focus on weld characteristics, including the location distribution, length, width, shape (such as butt welds, fillet welds, etc.) of the welds, as well as preliminary judgments on possible welding difficulty areas of the welds, such as welds in narrow spaces, welds connecting materials of different thicknesses, etc. This information provides key basic data for subsequent welding process planning.

[0057] The welded component data is scientifically classified according to the pre-set component structure characteristic variables. These variables may involve factors such as the component material type (such as carbon steel, aluminum alloy, etc.), shape complexity (simple plane components, complex curved surface components, etc.), load-bearing capacity requirements (main load-bearing components, secondary auxiliary components, etc.), and welding position and direction characteristics. Different types of welded components are determined by matching and classifying the welded component data with these variables. For example, all welded components made of carbon steel, with relatively simple shapes and used in the main load-bearing parts are classified into one category, while non-load-bearing components made of aluminum alloy and with complex curved shapes are classified into another category. This classification method helps to formulate personalized welding processes and parameter settings for different types of components in the future, thereby improving welding quality and efficiency.

[0058] A comprehensive traversal of the determined welding component types is carried out. For each type of component, the corresponding component welding requirements are configured in combination with its function and mechanical performance requirements in the frame structure, which includes selecting appropriate welding methods (such as manual arc welding, gas shielded welding, etc.), determining the specifications and models of welding materials (such as the diameter of the welding wire, the material of the welding rod, etc.), setting the preliminary range of welding parameters (current, voltage, welding speed, etc.) and the acceptance criteria for welding quality (such as weld strength requirements, appearance quality standards, etc.). Then, according to the pre-planned welding sequence, the welding requirements of various components are positively serialized and integrated in order, taking into full consideration factors such as thermal deformation control, assembly sequence and overall structural stability during welding, so as to determine the complete and reasonable welding requirements under the welding cycle of the frame. This welding requirement will serve as the core basis for subsequent welding process control, parameter optimization and quality monitoring, ensuring that the welding work of the entire electric tricycle frame can be carried out in an orderly manner, achieving high-quality and efficient welding production, improving the overall performance and reliability of the frame, and meeting the safety and durability requirements of the vehicle in actual use.

[0059] Embodiment 2, based on the same inventive concept as the welding quality monitoring method of the electric tricycle frame in the above embodiment, Figure 2 As shown, the present application provides a welding quality monitoring platform for an electric tricycle frame, and the platform and method embodiments in the present application embodiments are based on the same inventive concept. The platform includes: The welding requirement determination module 10 is used to perform a three-dimensional scan on the frame components of the electric tricycle to determine the welding requirements of the frame under the welding cycle of the whole vehicle, wherein the welding requirements of different welding structures are different.

[0060] The combined welding parameter sequence determination module 20 performs welding parameter analysis based on the welding requirement and in combination with the welding decision model to determine the combined welding parameter sequence under the periodic sequence.

[0061] The self-attention sequence configuration module 30 is used to locate the personalized monitoring features of each welding stage in the welding cycle according to the changes in welding requirements, and configure the self-attention sequence, wherein the personalized monitoring features are determined by the welding key parameters-expected welding features.

[0062] The welding data fragment determination module 40 is used to map the combined welding parameter sequence and the self-attention sequence, set the automatic monitoring module, perform welding control under the welding cycle by controlling the combined welding parameter sequence, trigger the personalized monitoring features based on the self-attention sequence based on the mapping relationship, perform personalized monitoring and data screening, and determine the welding data fragments.

[0063] The abnormal welding determination module 50 performs abnormal welding determination and feedback decision analysis on the welding data fragments based on the welding decision model, and reversely guides welding control and welding monitoring.

[0064] Furthermore, the self-attention sequence configuration module 30 further includes: A switching position determining unit is used to locate the welding position of a component with a changed welding requirement according to the welding cycle as a switching position for welding monitoring, and the welding position of the component is an initial position for entering the next welding stage.

[0065] A monitoring switching condition determination unit is used to map the switching position with the personalized monitoring feature to determine the monitoring switching condition.

[0066] A self-attention sequence representation unit is configured to identify the self-attention sequence based on the monitoring switching condition.

[0067] Furthermore, the welding data fragment determination module 40 further includes: A first welding parameter combination determining unit is configured to determine a first welding parameter combination based on the combined welding parameter sequence.

[0068] A welding data stream determination unit, wherein the welding data stream determination unit is used to perform a first stage of welding control through the first welding parameter combination, and determine a first self-attention feature based on a mapping relationship, control a monitoring device to perform synchronous welding monitoring, and determine a welding data stream, wherein monitoring is performed through collaborative vision, laser detection, and control of data streams.

[0069] A welding data stream screening unit is used to screen the welding data stream and determine the welding data fragments.

[0070] Furthermore, the abnormal welding determination module 50 further includes: A welding data fragment identification unit is used to identify the welding data fragments and integrate them to determine the first welding key parameter and the second welding feature.

[0071] The first monitoring result determination unit extracts key welding parameters based on the first welding parameter combination, performs parameter response judgment on the first welding key parameters, and determines a first monitoring result.

[0072] The second monitoring result determining unit determines the second monitoring result by determining the welding effect of the second welding feature based on the second expected welding feature.

[0073] A welding feedback strategy determination unit, wherein the welding feedback strategy determination unit is used to determine a welding feedback strategy according to the first monitoring result and the second monitoring result.

[0074] Furthermore, the welding feedback strategy determination unit further includes: The first feedback strategy determination unit is used to make a first feedback decision and determine a first feedback strategy if there is an abnormality in the first monitoring result or the second monitoring result.

[0075] The second feedback strategy determination unit is used to proofread and verify the first monitoring result and the second monitoring result, make a second feedback decision, and determine the second feedback strategy, wherein whether the first monitoring result and the second monitoring result are consistent is used as the proofreading and verification method, and the disturbance factor of the welding process under inconsistent proofreading and verification results is used as the feedback decision target.

[0076] A feedback strategy fusion unit is used to fuse the first feedback strategy and the second feedback strategy to determine the welding feedback strategy.

[0077] Furthermore, the feedback strategy fusion unit also includes: The first feedback strategy includes a first compensation strategy based on parameter response deviation and a second compensation strategy based on welding effect deviation, and the second feedback strategy is a compensation strategy based on disturbance factors, wherein the parameter response deviation is the difference between the welding key parameter in the first welding parameter combination and the first welding key parameter.

[0078] Furthermore, the welding requirement determination module 10 also includes: A welding component data positioning unit is used to determine the frame component scanning data and locate the welding component data by performing three-dimensional scanning, wherein the welding component data includes weld features.

[0079] A welding component class determination unit is configured to classify the welding component data based on preset variables of component structural characteristics to determine the welding component class.

[0080] A positive serialization integration unit is used to traverse the welding component class, configure component welding requirements, and perform positive serialization integration based on the welding sequence to determine the welding requirements under the frame vehicle welding cycle.

[0081] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0083] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for monitoring welding quality of an electric tricycle frame, characterized in that: The method comprises: Conduct 3D scanning of the frame components of the electric tricycle to determine the welding requirements of the frame under the welding cycle of the entire vehicle. Different welding structures have different welding requirements. Based on the welding requirements, a welding parameter analysis is performed in combination with a welding decision model to determine a combined welding parameter sequence under a periodic sequence; Based on the changes in welding requirements, the personalized monitoring features of each welding stage in the welding cycle are located, and a self-attention sequence is configured, in which the personalized monitoring features are determined by the welding key parameters-expected welding features; Mapping the combined welding parameter sequence and the self-attention sequence, setting an automatic monitoring module, performing welding control under a welding cycle by controlling the combined welding parameter sequence, triggering a personalized monitoring feature based on the self-attention sequence based on a mapping relationship, performing personalized monitoring and data screening, and determining welding data fragments; Based on the welding decision model, abnormal welding judgment and feedback decision analysis are performed on the welding data fragments to reversely guide welding control and welding monitoring.

2. The welding quality monitoring method of the electric tricycle frame according to claim 1, characterized in that: The configuration self-attention sequence includes: For the welding cycle, the welding position of the component with the changed welding demand is located as the switching position of the welding monitoring, and the welding position of the component is the initial position for entering the next welding stage; Mapping the switching position with the personalized monitoring feature to determine a monitoring switching condition; Based on the monitoring switching condition, the self-attention sequence is identified.

3. The welding quality monitoring method of the electric tricycle frame according to claim 1, characterized in that: The determining of welding data fragments comprises: Based on the combined welding parameter sequence, determining a first welding parameter combination; Through the first welding parameter combination, the first stage of welding control is performed, and the first self-attention feature based on the mapping relationship is determined, and the monitoring device is controlled to perform synchronous welding monitoring and determine the welding data stream, wherein the monitoring is performed through coordinated vision, laser detection and control data stream; The welding data stream is screened to determine the welding data fragments.

4. The welding quality monitoring method of the electric tricycle frame according to claim 3, characterized in that: Performing abnormal welding judgment and feedback decision analysis on the welding data fragments, including: Identify the welding data fragments, integrate them to determine the first welding key parameter and the second welding feature; Based on the first welding parameter combination, extracting key welding parameters, performing parameter response determination on the first key welding parameters, and determining a first monitoring result; Based on the second expected welding feature, performing welding effect judgment on the second welding feature to determine a second monitoring result; A welding feedback strategy is determined according to the first monitoring result and the second monitoring result.

5. The welding quality monitoring method of the electric tricycle frame according to claim 4, characterized in that: Determining a welding feedback strategy according to the first monitoring result and the second monitoring result includes: If the first monitoring result or the second monitoring result is abnormal, make a first feedback decision and determine a first feedback strategy; Proofread and verify the first monitoring result and the second monitoring result, make a second feedback decision, and determine a second feedback strategy, wherein whether the first monitoring result and the second monitoring result are consistent is used as a proofreading and verification method, and the disturbance factor of the welding process under inconsistent proofreading and verification results is used as a feedback decision target; The first feedback strategy and the second feedback strategy are integrated to determine the welding feedback strategy.

6. The welding quality monitoring method of the electric tricycle frame according to claim 5, characterized in that: The first feedback strategy includes a first compensation strategy based on parameter response deviation and a second compensation strategy based on welding effect deviation, and the second feedback strategy is a compensation strategy based on disturbance factors, wherein the parameter response deviation is the difference between the welding key parameter in the first welding parameter combination and the first welding key parameter.

7. The welding quality monitoring method of the electric tricycle frame according to claim 1, characterized in that: Determine the welding requirements for the vehicle frame welding cycle, including: Determine the scanning data of the vehicle frame component by performing three-dimensional scanning, and locate the welding component data, wherein the welding component data includes weld features; Classifying the welding component data based on preset variables of component structural characteristics to determine the welding component class; The welding component classes are traversed, component welding requirements are configured, and positive serialization integration is performed based on the welding sequence to determine the welding requirements under the frame vehicle welding cycle.

8. The welding quality monitoring platform of the electric tricycle frame is characterized by: The platform is used to implement the welding quality monitoring method of the electric tricycle frame according to any one of claims 1 to 7, and the platform comprises: The welding requirement determination module is used to perform three-dimensional scanning on the frame components of the electric tricycle to determine the welding requirements under the welding cycle of the frame. The welding requirements of different welding structures are different. A combined welding parameter sequence determination module, based on the welding requirements and in combination with a welding decision model, performs welding parameter analysis to determine a combined welding parameter sequence under a periodic sequence; A self-attention sequence configuration module is used to locate the personalized monitoring features of each welding stage in the welding cycle according to the changes in welding requirements, and configure the self-attention sequence, wherein the personalized monitoring features are determined by the welding key parameters-expected welding features; A welding data fragment determination module is used to map the combined welding parameter sequence and the self-attention sequence, set an automatic monitoring module, perform welding control under a welding cycle by controlling the combined welding parameter sequence, trigger a personalized monitoring feature based on the self-attention sequence based on the mapping relationship, perform personalized monitoring and data screening, and determine welding data fragments; The abnormal welding determination module performs abnormal welding determination and feedback decision analysis on the welding data fragments based on the welding decision model, and reversely guides welding control and welding monitoring.

Images