A welding control optimization method and system for welded pipe manufacturing

Abstract

The application provides a welding control optimization method and system for welded pipe manufacturing, and relates to the technical field of welding control. The method comprises the following steps: collecting welding process record data of a plurality of welded pipes, identifying a plurality of state sampling sections, constructing a state transition path of each set of welding process record data, dividing the plurality of sets of welding process record data into normal samples and abnormal samples, constructing a state evolution graph based on the plurality of normal samples, identifying a deviation origin of each abnormal sample, constructing a plurality of deviation mode nodes based on the plurality of deviation origins, determining a plurality of candidate matching nodes of each deviation mode node, constructing a target evolution domain based on the plurality of normal samples, extracting a state regression anchor point of each deviation origin, constructing an abnormal repair database, performing abnormal monitoring on the production process of a target welded pipe through the state evolution graph, and performing welding control optimization based on the abnormal repair database. The application improves the stability of the welding process control of the welded pipe.

Description

Technical Field

[0001] This invention relates to the field of welding control technology, and specifically to a welding control optimization method and system for welded pipe manufacturing. Background Technology

[0002] In the manufacturing process of welded pipes, the welding process is a crucial step affecting the quality of the welded pipe and the stability of the weld seam. Under continuous, assembly line production conditions, various welding parameters, such as welding current, voltage, and welding speed, evolve gradually with the production process, and the parameter states and their fluctuation characteristics differ at different production stages. If the parameter states deviate during the welding process and are not identified and reasonably adjusted in time, it can easily have an adverse impact on the weld seam quality and production stability.

[0003] Many welding anomaly detection methods rely on fixed thresholds or local parameter changes for judgment, typically focusing on identifying and alarming instantaneous anomalies. While these methods can reflect significant parameter anomalies to some extent, they fail to fully consider the stage characteristics of the welding process and the evolution of multiple parameter states over time. Consequently, they struggle to accurately characterize gradual shifts or state evolution anomalies that occur during welding, making it difficult to provide a continuous and targeted basis for subsequent parameter control.

[0004] Furthermore, after welding anomalies occur, some techniques rely on empirical rules or simple feedback strategies to adjust parameters, lacking utilization of parameter evolution paths during historical normal production processes. This can easily lead to mismatches in control amplitude or inconsistencies with the current production stage. Therefore, it is necessary to propose a welding control optimization method that combines the characteristics of welding process stages and the evolution laws of normal states to improve the stability and rationality of control in the welded pipe welding process. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a welding control optimization method and system for welded pipe manufacturing. By modeling and analyzing the multi-parameter states and their evolution relationships in the welding process, it achieves staged analysis and control optimization of welding deviation states, thereby improving the stability and control rationality of the welded pipe welding process.

[0006] To achieve the above objectives, the first aspect of the present invention provides a welding control optimization method for welded pipe manufacturing, comprising: Welding process recording data of multiple welded pipes is collected. The welding process recording data includes welding time sequence data of multiple welding parameters and welding quality data of welded pipes. Multiple state sampling segments of each group of welding time sequence data are identified. For each state sampling segment, identify the parameter association state between any two welding parameters, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data; Based on the welding quality data, multiple sets of welding process records are divided into normal samples and abnormal samples. A state evolution map is constructed based on the state transition paths of multiple normal samples. The deviation from the origin of each abnormal sample is identified based on the state evolution map. Multiple deviation pattern nodes are constructed based on multiple deviations from the origin, and multiple candidate matching nodes are determined for each deviation pattern node based on multiple normal samples. Based on multiple normal samples, multiple closed nodes in the state evolution map are determined and a target evolution domain is constructed. Based on the target evolution domain, each state regression anchor point deviating from the origin is extracted. An anomaly repair database is constructed based on multiple state regression anchor points. Anomalies in the production process of the target welded pipe are monitored through state evolution maps. After identifying production deviations, welding control is optimized based on the anomaly repair database.

[0007] Preferably, a state evolution graph is constructed based on the state transition paths of multiple normal samples, and the deviation from the origin of each abnormal sample is identified based on the state evolution graph, including: Using multiple state signatures included in multiple normal samples as graph nodes, a state evolution graph for multiple normal samples is constructed, where each graph node in the state evolution graph corresponds to a unique state signature. Based on the state transition paths of multiple normal samples, the node frequency of each graph node in the state evolution graph is counted, and the transition frequency of each directed transition pair formed by any two graph nodes in the state evolution graph is counted. Identify the abnormal state of each abnormal sample in the state transition path based on multiple node frequencies and transition frequencies, including marking path nodes in the state transition path with node frequencies lower than the node abnormality threshold as node frequency abnormal states, marking path nodes with transition frequencies lower than the transition abnormality threshold as transition frequency abnormal states; and marking multiple path nodes belonging to abnormal states as abnormal nodes. For each anomalous sample, identify multiple local anomalous paths in the state transition path, and mark the starting node of the first local anomalous path as the deviation point from the origin of the anomalous sample.

[0008] Preferably, constructing multiple deviation pattern nodes based on multiple deviations from the origin, and determining multiple candidate matching nodes for each deviation pattern node based on multiple normal samples includes: Clustering is performed on multiple deviation points based on the state signatures of the deviation points to generate multiple deviation patterns corresponding to the deviation points. A signature mean vector is constructed based on the multiple deviation points included in the deviation patterns, and the signature mean vector is used as the deviation pattern node of the deviation pattern. The process involves matching the deviant pattern node with multiple path nodes contained in the normal sample. This includes calculating the matching index between the deviant pattern node and multiple path nodes based on the signature mean vector and the state signature of the path node, and then selecting multiple candidate matching nodes for the deviant pattern node from the multiple path nodes based on the matching index.

[0009] Preferably, determining multiple closed nodes in the state evolution graph based on multiple normal samples and constructing the target evolution domain includes: Extract the local closed path of each normal sample according to the preset closed time period, and mark multiple path nodes in the local closed path as closed nodes; An initial evolutionary domain is constructed based on multiple closed nodes. Based on the initial evolutionary domain, the remaining graph nodes are expanded by reverse traversal along the state evolution graph. During the traversal, for any candidate node, if the transition frequency of the current candidate node to any node in the initial evolutionary domain is higher than the preset evolutionary transition threshold, the current candidate node is added to the initial evolutionary domain. Otherwise, the reverse traversal expansion of the current candidate node is stopped. When there are no candidate nodes that can be expanded by reverse traversal, the traversal is stopped, and the target evolutionary domain after expanding the initial evolutionary domain is obtained.

[0010] Preferably, extracting each state regression anchor point deviating from the origin based on the target evolution domain includes: Based on the target evolution domain and the state transition path to which the candidate matching node belongs, identify the local state anchor point corresponding to each candidate matching node; Determine the multiple production stages of welded pipe production and identify the production stage to which multiple path nodes in each state transition path belong; Based on the production stage to which the path node belongs, determine the candidate stage range for each deviation from the origin. Based on the candidate stage range, select multiple stage matching nodes from the multiple candidate matching nodes contained in the deviation pattern node to which the deviation from the origin belongs. Extract the state signature difference and sequence position difference between the deviation from the origin and the local state anchor point of each stage matching node. Calculate the evolution matching parameters corresponding to the deviation from the origin and multiple local state anchor points based on the state signature difference and sequence position difference. Select the state regression anchor point corresponding to the deviation from the origin from multiple local state anchor points based on the evolution matching parameters.

[0011] Preferably, the production process of the target welded pipe is monitored for anomalies using a state evolution map, and welding control is optimized based on an anomaly repair database after identifying production deviations. Welding monitoring data of the target welded pipe is collected, and real-time state signatures corresponding to the welding monitoring data in multiple state sampling segments are extracted to construct the real-time state transition path of the target welded pipe. Based on the state evolution graph, the real-time deviation point of the real-time state transition path is identified. The real-time deviation point is matched with multiple deviation pattern nodes to determine the target deviation pattern node corresponding to the real-time deviation point. Based on the anomaly repair database, the state regression anchor points corresponding to the target deviation pattern nodes in multiple production stages are determined. The optimization anchor point is determined according to the production stage to which the real-time deviation point belongs. The target optimization strategy is generated based on the optimization anchor point. The welding control of the target welded pipe is optimized based on the target optimization strategy.

[0012] A second aspect of the present invention provides a welding control optimization system for welded pipe manufacturing, used to implement the aforementioned welding control optimization method for welded pipe manufacturing, comprising: The welding data acquisition module is used to collect welding process record data of multiple welded pipes. The welding process record data includes welding time sequence data of multiple welding parameters and welding quality data of welded pipes, and identifies multiple state sampling segments of each group of welding time sequence data. The state transition analysis module is used to identify the parameter association state between any two welding parameters in each state sampling segment, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data. The sample deviation identification module is used to divide multiple sets of welding process record data into normal samples and abnormal samples based on welding quality data, construct a state evolution map based on the state transition paths of multiple normal samples, and identify the deviation point from the origin of each abnormal sample based on the state evolution map. The matching node generation module is used to construct multiple deviation pattern nodes based on multiple deviations from the origin, and to determine multiple candidate matching nodes for each deviation pattern node based on multiple normal samples. The state regression analysis module is used to determine multiple closed nodes in the state evolution map based on multiple normal samples and construct the target evolution domain. It also extracts the state regression anchor point for each deviation from the origin based on the target evolution domain. The welding control optimization module is used to build an anomaly repair database based on multiple state regression anchor points, monitor the production process of the target welded pipe through the state evolution map, and optimize welding control based on the anomaly repair database after identifying the production deviation state.

[0013] The present invention has the following beneficial effects: This invention focuses on the state evolution characteristics of the welded pipe welding process. It constructs a state evolution map based on normal samples and constrains the stable evolution path of the final welding state by introducing closed nodes and a target evolution domain. This allows the state map to characterize the reliable evolution structure of the welding process from the initial state to the steady-state endpoint. Furthermore, by identifying deviations from the origin and matching deviation patterns in real-time welding state transition paths, abnormal states are mapped to deviation pattern nodes with clear evolutionary semantics. Combined with state regression anchor points corresponding to different production stages, an optimized reference state matching the current stage is selected from the target evolution domain, thereby generating a targeted welding control optimization strategy. By utilizing the consistency and reachability characteristics of the welding state in the evolutionary structure, the anomaly repair process is constrained, avoiding the uncertainty caused by relying solely on instantaneous states or complete historical paths for control decisions. This gives welding control adjustments clear evolutionary direction and stage adaptability, helping to improve the stability of welding process control and the consistency of welding quality. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a welding control optimization method for welded pipe manufacturing, provided as an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of a welding control optimization system for welded pipe manufacturing, provided as an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0017] Please see Figure 1 The first aspect of this invention provides a welding control optimization method for welded pipe manufacturing, the method comprising: Step S1: Collect welding process recording data of multiple welded pipes. The welding process recording data includes welding timing data of multiple welding parameters and welding quality data of welded pipes. Identify multiple state sampling segments of each group of welding timing data.

[0018] In this embodiment, taking the ERW high-frequency resistance welded steel pipe production line as an example, each welded pipe corresponds to a complete set of welding process record data during the production process. This record data includes, but is not limited to, welding timing data of welding parameters such as welding current, welding voltage, welding speed, and extrusion pressure. The above welding parameters are continuously collected at a preset sampling frequency during the welding process and are associated with and stored with the welding quality inspection results of the corresponding welded pipe. The welding quality data may include qualified / unqualified identification information such as weld formation quality and weld strength test results, which are used to characterize the final quality state corresponding to the welding process of the welded pipe.

[0019] It is worth noting that, to facilitate the characterization of the evolution of multiple parameters during the welding process, the continuously acquired welding time-series data is divided into multiple local analysis segments according to time sequence. Each local analysis segment corresponds to a continuous time window. Within this time window, the statistical characteristics of each welding parameter remain relatively stable, and their variation amplitude is within a preset allowable range, constituting a basic state sampling segment. Those skilled in the art can reasonably set the length of the time window and the preset allowable range of different welding parameters according to the actual welding process of the welded pipe. For example, the variance value of the parameters within the set time window can be calculated. If the variance value exceeds the corresponding preset allowable range, the length of the time window can be appropriately reduced so that the variation amplitude of multiple welding parameters within different windows is within a suitable range. The finally determined multiple state sampling segments serve as the basic data units for subsequent welding state analysis and evolution modeling.

[0020] Step S2: Identify the parameter association state between any two welding parameters for each state sampling segment, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data.

[0021] In this embodiment, for any given state sampling segment, the parameter correlation state between any two welding parameters is further analyzed. Specifically, the correlation coefficient between the time series of two welding parameters within the state sampling segment can be calculated to quantify the changing trend and coupling degree between the two parameters. For example, the Pearson correlation coefficient between the two time series is calculated. If the Pearson correlation coefficient is positive and its absolute value is greater than a preset correlation threshold, the parameter correlation state between the two welding parameters is recorded as positively correlated; if the Pearson correlation coefficient is negative and its absolute value is greater than the preset correlation threshold, the parameter correlation state between the two welding parameters is recorded as negatively correlated; otherwise, the parameter correlation state between the two welding parameters is recorded as decoupled.

[0022] It is worth noting that those skilled in the art can use other suitable indicators to measure the degree of interaction between two welding parameters, as well as the coupling or unstable state between parameters. Based on the obtained parameter correlation state between any two welding parameters, different correlation states can be discretized and encoded, such as positive correlation, negative correlation, and decoupling, which are sequentially denoted as 1, 2, and 3. The discretized parameter correlation states of all welding parameter pairs in each state sampling segment are combined to form the state signature vector of that segment. Each state signature vector uses the same parameter template, meaning that the same position in multiple state signature vectors corresponds to the discretized parameter correlation state combination of the same welding parameter pair. The state signatures of continuous state sampling segments of the same welded pipe are combined in chronological order to obtain the state transition path of each group of welding process recorded data.

[0023] It should be understood that the above-described correlation analysis and discretization processing are merely exemplary implementation methods. Those skilled in the art can select or design other suitable parameter correlation indicators and state discretization methods according to actual production processes and data characteristics to construct corresponding state signatures.

[0024] Step S3: Divide multiple sets of welding process record data into normal samples and abnormal samples according to the welding quality data. Construct a state evolution map based on the state transition paths of multiple normal samples. Identify the deviation from the origin of each abnormal sample based on the state evolution map.

[0025] In this embodiment, the welding process record data corresponding to each welded pipe is associated with its weld quality pass / fail indicator, and multiple sets of welding process record data are divided into normal samples and abnormal samples. By analyzing the state evolution patterns of multiple normal samples with qualified weld quality, a state evolution map is constructed based on the state transition paths of multiple normal samples to reflect the evolutionary relationships between various stages of the normal production process. Simultaneously, based on the state evolution map, the state transition paths of each abnormal sample are compared and analyzed to identify the node in the state transition path where the abnormal sample first deviates from the normal trajectory, i.e., the deviation from the origin.

[0026] As an exemplary implementation, a state evolution graph is constructed based on the state transition paths of multiple normal samples. The deviation from the origin for each abnormal sample is identified based on this graph, which may include the following: For the construction of the state evolution map, multiple state signatures included by multiple normal samples are used as map nodes. By integrating the state signatures of all normal samples, a state evolution map about multiple normal samples is constructed. Each map node in the state evolution map corresponds to a unique state signature, which is used to characterize the typical combination of parameter states and their evolution law in the normal production process. The nodes in the state evolution map not only contain the correlation information between parameters, but also reflect the possible paths of state evolution through the transition relationship between nodes.

[0027] Based on the state transition paths of multiple normal samples, the node frequency of each graph node in the state evolution graph is statistically analyzed, and the transition frequency of each directed transition pair formed by any two graph nodes in the state evolution graph is statistically analyzed. The node frequency of a graph node represents the percentage of times that state signature appears in all normal samples, thus measuring the representativeness of the node in the normal production process. The transition frequency of a directed transition pair represents the percentage of times that transition relationship appears in all normal samples. For example, the transition frequency from node A to node B is the percentage of times a transition from node A to node B occurs in all transition processes starting from node A, reflecting the typicality and stability of state transitions.

[0028] The abnormal state of each anomalous sample in the state transition path is identified based on multiple node frequencies and transition frequencies. These abnormal states include node frequency abnormalities and transition frequency abnormalities. Specifically, path nodes in the state transition path whose node frequency is below a node abnormality threshold are marked as node frequency abnormalities, and path nodes whose transition frequency is below a transition abnormality threshold are marked as transition frequency abnormalities. The transition frequency of a path node specifically refers to the transition frequency between the current path node and the next path node in the state transition path. Both types of abnormal state nodes are marked as anomalous nodes.

[0029] Finally, it is determined that for each anomalous sample, multiple consecutive anomalous nodes in the state transition path constitute a local anomalous path. To ensure the representativeness of the local anomalous paths, the number of anomalous nodes constituting the local anomalous path can be set to be no less than a path length threshold. For these local anomalous paths, the starting node of the first local anomalous path is taken as the deviation point from the origin of the anomalous sample, representing the position where the anomalous sample first deviates from the normal state evolution trajectory.

[0030] Step S4: Construct multiple deviation pattern nodes based on multiple deviations from the origin, and determine multiple candidate matching nodes for each deviation pattern node based on multiple normal samples.

[0031] In this embodiment, multiple deviation patterns are identified by clustering the deviation points of multiple abnormal samples, and a deviation pattern node corresponding to each deviation pattern is constructed. Then, nodes with similar features to the deviation pattern node are searched in normal samples to form a candidate matching node set for each deviation pattern node, which includes multiple candidate matching nodes for the deviation pattern.

[0032] As an exemplary implementation, the process of constructing multiple deviation pattern nodes and determining multiple candidate matching nodes for each deviation pattern node specifically includes: To extract typical features of different deviation types, multiple deviation points are clustered based on their state signatures. The distance between any two deviation points is calculated using the state signatures, for example, by calculating the Euclidean distance between two state signatures. Similar deviation points in the state signature feature space are then grouped together using a clustering algorithm, such as K-means clustering, to generate multiple deviation patterns corresponding to the deviation points. Those skilled in the art can choose appropriate clustering algorithms according to actual needs; this embodiment does not specifically limit them. Each resulting deviation pattern represents a typical deviation behavior, reflecting an abnormal trend or deviation mode of parameter state combinations during the welding process.

[0033] For each deviation pattern, statistical analysis is performed on the state signatures of the multiple deviation points contained therein, and the mean vector of the state signatures of each deviation point is calculated to form a deviation pattern node, which is used to represent the typical state of the deviation pattern.

[0034] Furthermore, the deviation pattern nodes are matched with multiple path nodes contained in the normal samples. During the matching process, the matching index between the deviation pattern node and multiple path nodes can be calculated based on the mean signature vector of the deviation pattern node and the state signature vector of the path nodes. The matching index can comprehensively consider quantifiable feature parameters such as the similarity and distance of the node state signatures. In this embodiment, Euclidean distance is used as an example to calculate the Euclidean distance between two vectors as the matching index.

[0035] By comparing the matching indices of multiple path nodes, several path nodes that are closest to the deviation pattern node are selected from the multiple path nodes of the normal sample. For example, path nodes whose matching indices are less than a preset distance threshold are marked as candidate matching nodes for the deviation pattern node. The aforementioned candidate matching nodes are used to describe the potential starting point of the sample state evolution that may return to the normal production process after a similar deviation state occurs during normal production.

[0036] Step S5: Based on multiple normal samples, determine multiple closed nodes in the state evolution map and construct the target evolution domain. Extract each state regression anchor point that deviates from the origin according to the target evolution domain.

[0037] In this embodiment, based on data from multiple normal samples, closed nodes in the state evolution graph are identified. These closed nodes represent key states to which the state evolution path will eventually revert during normal production. An initial evolution domain is constructed based on these closed nodes. Multiple nodes with a high probability of reaching the initial evolution domain are identified to form a target evolution domain, representing the range of state space to which normal samples may revert to a steady state during production. Furthermore, based on the target evolution domain, state regression anchor points corresponding to each deviation from the origin are extracted. These are the key nodes where the deviated nodes revert to a stable and controllable state in normal samples, providing a reference direction for anomaly repair.

[0038] As an exemplary implementation, the above-mentioned method of determining multiple closed nodes in the state evolution graph based on multiple normal samples and constructing a target evolution domain includes the following specific implementation process: The local closed path of each normal sample is extracted according to the preset closed time period. The preset closed time period is a period of time before the end of welding, which represents the ideal steady state when the welded pipe is completed, that is, a period when the welding work is about to end. The specific closed time period length can be reasonably set by those skilled in the art according to the actual welding process. Multiple path nodes in this period constitute a local closed path, and multiple path nodes are marked as closed nodes.

[0039] An initial evolutionary domain is constructed based on multiple closed nodes, which serve as starting nodes. The remaining nodes in the graph are then expanded in reverse traversal along the state evolution graph based on this initial evolutionary domain. Specifically, diffusion occurs based on these closed nodes. When any candidate node is encountered, the transition frequency between that candidate node and any node in the initial evolutionary domain in the current state is determined according to the state evolution graph. If the transition frequency between the candidate node and any node in the initial evolutionary domain is higher than a preset evolutionary transition threshold, the candidate node is added to the initial evolutionary domain, and the reverse traversal continues from that node as a new starting point. If the transition frequency does not meet the threshold condition, the expansion of that candidate node is stopped to prevent low-probability paths from affecting the stability of the evolutionary domain.

[0040] Repeat the reverse traversal expansion process described above until no candidate nodes remain for reverse traversal expansion. This yields the target evolutionary domain after reverse traversal expansion of the initial evolutionary domain. By traversing the state graph backwards from the closed node, we can identify which preceding states will eventually revert to the closed node during normal production. This establishes stable evolutionary paths. By setting an evolutionary transition threshold, only high-probability paths are included, ensuring that the target evolutionary domain contains only truly feasible state sequences and avoiding deviations from anomalies or occasional fluctuations that could affect the reference.

[0041] In this embodiment, the target evolution domain is a stable state subspace constructed based on the state evolution graphs of multiple normal samples. It is used to characterize the reliability and controllability of each node on the evolution path during the welding process. By starting from the steady-state node at the end of the welding stage and expanding backward along the state graph, each node in the target evolution domain satisfies the following characteristics: once at this node, its own state information can uniquely determine its subsequent evolution path, thereby enabling a stable regression to the steady-state node at the end of the welding stage without relying on the preceding path combination or subsequent selection of the node. This avoids path ambiguities or uncertainties that may exist during the welding process.

[0042] As an exemplary implementation, each state regression anchor point deviating from the origin is extracted based on the target evolution domain, specifically including: Based on the state transition paths of the target evolution domain and the candidate matching nodes, the local state anchor point corresponding to each candidate matching node is identified. The local state anchor point is the first path node belonging to the target evolution domain in the state transition path of the candidate matching node, which represents the reference state to which the node can reliably return to the steady state during normal production.

[0043] For assembly line operation scenarios, the entire welding process of welded pipe production can be divided into several stages or windows of equal length to distinguish the corresponding welding progress. Based on the multiple production stages of welded pipe production, the production stage to which multiple path nodes in each state transition path belong can be identified.

[0044] Based on this, the range of candidate stages for each deviation from the origin is determined according to the production stage to which the path node belongs. The candidate stages can be based on the production stage where the deviation from the origin is located and expanded by a preset expansion amount, that is, expanding the range forward and backward by a preset expansion amount, for example, expanding the stage duration by 10% in both directions, to construct the range of candidate stages for deviation from the origin. Based on the range of candidate stages, multiple stage matching nodes for deviation from the origin are selected from multiple candidate matching nodes contained in the deviation mode node to which the deviation from the origin belongs. These stage matching nodes represent the reference starting point of the feasible regression path of the deviation from the origin in the corresponding production stage.

[0045] Then, the state signature difference and sequence position difference between the deviation from the origin and the local state anchor point of each stage matching node are extracted. The state signature difference can be the distance between the state signature vectors of the deviation from the origin and the local state anchor point, reflecting the difference between the deviation node and the local state anchor point in the associated states of multiple welding parameters. The larger the difference, the greater the adjustment required. In this embodiment, the state signature difference is represented by calculating the Euclidean distance between the state signature vectors. The sequence position difference is used to reflect the relative position difference between the deviation from the origin and the local state anchor point in the time series of the welding process. Specifically, it is the number of production stages between the two, and is normalized based on the total number of production stages. The ratio between the number of production stages between them and the total number of production stages is calculated as the sequence position difference.

[0046] Based on the differences in state signatures and sequence positions, the differences in state signatures are corrected by adjusting the differences in sequence positions. Specifically, the product of the differences in state signatures and sequence positions is calculated to obtain the evolution matching parameters corresponding to the deviation from the origin and multiple local state anchor points. Finally, based on the evolution matching parameters, the local state anchor point with the smallest evolution matching parameters is selected as the state regression anchor point corresponding to the deviation from the origin. The state regression anchor point represents the node with the smallest combined difference in distance from the deviation from the origin in the production stage and the difference in the associated states of multiple welding parameters. This eliminates the need for significant adjustments to welding parameters, and the adjustment target does not need to span multiple production stages, which conforms to the staged unidirectional advancement process of welded pipe welding.

[0047] Step S6: Construct an anomaly repair database based on multiple state regression anchor points, monitor the production process of the target welded pipe through the state evolution map, and optimize welding control based on the anomaly repair database after identifying the production deviation state.

[0048] In this embodiment, by organizing the state regression anchor points corresponding to each deviation mode node at different production stages, an anomaly repair database that can be used for online reference is formed. During the welding production process of the target welded pipe, the state of welding parameters is monitored in real time through the state evolution map. When a deviation state is detected, a corresponding repair strategy can be generated based on the anchor point information provided by the anomaly repair database. The welding parameters are then optimized and adjusted according to the repair strategy to achieve welding control optimization and real-time assurance of welded pipe quality.

[0049] As one exemplary implementation, anomaly monitoring of the target welded pipe production process is performed using a state evolution graph. After identifying production deviations, welding control optimization is performed based on an anomaly repair database, including: Welding monitoring data of the target welded pipe is collected, and real-time state signatures corresponding to multiple state sampling segments are extracted to construct the real-time state transition path of the target welded pipe. In this process, the welding monitoring data is segmented according to the aforementioned multiple state sampling segments, and the parameter correlation states between different welding parameters in each state sampling segment are extracted. This constructs the real-time state signature corresponding to each state sampling segment, and the real-time state transition path of the target welded pipe is obtained by temporal combination.

[0050] Based on the state evolution graph, the real-time deviation from the origin of the real-time state transition path is identified. Referring to the aforementioned process of identifying the deviation from the origin of abnormal samples, for the extraction of local abnormal paths, in the real-time monitoring process, an appropriate path length can be used as the extraction length of the local abnormal path according to the actual monitoring accuracy requirements. After determining the real-time deviation from the origin, the real-time deviation from the origin is matched with multiple deviation pattern nodes. Specifically, the Euclidean distance between the real-time state signature of the real-time deviation from the origin and the signature mean vector of the deviation pattern node is calculated as the distance value, and the deviation pattern node with the smallest distance value is selected as the target deviation pattern node corresponding to the real-time deviation from the origin.

[0051] Based on the anomaly repair database, the state regression anchor points corresponding to the target deviation pattern node in multiple production stages are determined. The anomaly repair database includes the state regression anchor points for each deviation pattern node in different production stages, specifically including the state signatures of the state regression anchor points, which include the parameter correlation states between different welding parameters. Based on the production stage to which the real-time deviation origin belongs, the corresponding optimization anchor point is determined as the target reference state for welding control optimization of the real-time deviation origin. Using the optimization anchor point as the target, the state differences between the real-time deviation origin and the optimization anchor point are analyzed to generate a target optimization strategy for the current welding anomaly. This includes the adjustment direction of relevant welding parameters such as welding current, welding voltage, and welding speed. That is, based on the relationship between the current state and the optimization anchor point, the challenge direction and adjustment range of multiple welding parameters are determined. The adjustment range can be reasonably set according to actual production, for example, each adjustment should not exceed a specific threshold to achieve small-range adjustments to the welding parameters. The target optimization strategy is used to indicate how to transform the real-time deviation from the origin state to the multi-parameter correlation state represented by the optimization anchor point, so that the real-time welding state of the target welded pipe changes along a reliable evolution direction and gradually returns to a stable welding state, thereby improving welding quality and the stability of the production process.

[0052] It is worth noting that the aforementioned thresholds are merely triggering conditions, and their specific values ​​are not key parameters for achieving the technical effects of this invention. Variations within a reasonable range will not affect the implementation of the technical solution of this invention. The innovation of this invention lies not in the specific numerical settings of the different thresholds, but in the processing logic and overall technical solution based on the different thresholds. Different thresholds can be preset or dynamically adjusted according to actual application scenarios, system operating states, or empirical data. Those skilled in the art can make reasonable selections or determinations based on specific needs.

[0053] Please see Figure 2 The second aspect of this invention provides a welding control optimization system for welded pipe manufacturing, comprising: The welding data acquisition module is used to collect welding process record data of multiple welded pipes. The welding process record data includes welding time sequence data of multiple welding parameters and welding quality data of welded pipes, and identifies multiple state sampling segments of each group of welding time sequence data. The state transition analysis module is used to identify the parameter association state between any two welding parameters in each state sampling segment, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data. The sample deviation identification module is used to divide multiple sets of welding process record data into normal samples and abnormal samples based on welding quality data, construct a state evolution map based on the state transition paths of multiple normal samples, and identify the deviation point from the origin of each abnormal sample based on the state evolution map. The matching node generation module is used to construct multiple deviation pattern nodes based on multiple deviations from the origin, and to determine multiple candidate matching nodes for each deviation pattern node based on multiple normal samples. The state regression analysis module is used to determine multiple closed nodes in the state evolution map based on multiple normal samples and construct the target evolution domain. It also extracts the state regression anchor point for each deviation from the origin based on the target evolution domain. The welding control optimization module is used to build an anomaly repair database based on multiple state regression anchor points, monitor the production process of the target welded pipe through the state evolution map, and optimize welding control based on the anomaly repair database after identifying the production deviation state.

[0054] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A welding control optimization method for welded pipe manufacturing, characterized in that, include: Welding process recording data of multiple welded pipes is collected. The welding process recording data includes welding time sequence data of multiple welding parameters and welding quality data of welded pipes. Multiple state sampling segments of each group of welding time sequence data are identified. For each state sampling segment, identify the parameter association state between any two welding parameters, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data; Based on the welding quality data, multiple sets of welding process records are divided into normal samples and abnormal samples. A state evolution map is constructed based on the state transition paths of multiple normal samples. The deviation from the origin of each abnormal sample is identified based on the state evolution map. Multiple deviation pattern nodes are constructed based on multiple deviations from the origin, and multiple candidate matching nodes are determined for each deviation pattern node based on multiple normal samples. Based on multiple normal samples, multiple closed nodes in the state evolution map are determined and a target evolution domain is constructed. Based on the target evolution domain, each state regression anchor point deviating from the origin is extracted. An anomaly repair database is constructed based on multiple state regression anchor points. Anomalies in the production process of the target welded pipe are monitored through state evolution maps. After identifying production deviations, welding control is optimized based on the anomaly repair database.

2. The welding control optimization method for welded pipe manufacturing according to claim 1, characterized in that, A state evolution graph is constructed based on the state transition paths of multiple normal samples. The deviation from the origin of each abnormal sample is identified based on the state evolution graph, including: Using multiple state signatures included in multiple normal samples as graph nodes, a state evolution graph for multiple normal samples is constructed, where each graph node in the state evolution graph corresponds to a unique state signature. Based on the state transition paths of multiple normal samples, the node frequency of each graph node in the state evolution graph is counted, and the transition frequency of each directed transition pair formed by any two graph nodes in the state evolution graph is counted. Identify the abnormal state of each abnormal sample in the state transition path based on multiple node frequencies and transition frequencies, including marking path nodes in the state transition path with node frequencies lower than the node abnormality threshold as node frequency abnormal states, marking path nodes with transition frequencies lower than the transition abnormality threshold as transition frequency abnormal states; and marking multiple path nodes belonging to abnormal states as abnormal nodes. For each anomalous sample, identify multiple local anomalous paths in the state transition path, and mark the starting node of the first local anomalous path as the deviation point from the origin of the anomalous sample.

3. The welding control optimization method for welded pipe manufacturing according to claim 2, characterized in that, Multiple deviation pattern nodes are constructed based on multiple deviations from the origin. Multiple candidate matching nodes for each deviation pattern node are determined based on multiple normal samples, including: Clustering is performed on multiple deviation points based on the state signatures of the deviation points to generate multiple deviation patterns corresponding to the deviation points. A signature mean vector is constructed based on the multiple deviation points included in the deviation patterns, and the signature mean vector is used as the deviation pattern node of the deviation pattern. The process involves matching the deviant pattern node with multiple path nodes contained in the normal sample. This includes calculating the matching index between the deviant pattern node and multiple path nodes based on the signature mean vector and the state signature of the path node, and then selecting multiple candidate matching nodes for the deviant pattern node from the multiple path nodes based on the matching index.

4. The welding control optimization method for welded pipe manufacturing according to claim 3, characterized in that, Based on multiple normal samples, multiple closed nodes in the state evolution graph are identified, and the target evolution domain is constructed, including: Extract the local closed path of each normal sample according to the preset closed time period, and mark multiple path nodes in the local closed path as closed nodes; An initial evolutionary domain is constructed based on multiple closed nodes. Based on the initial evolutionary domain, the remaining graph nodes are expanded by reverse traversal along the state evolution graph. During the traversal, for any candidate node, if the transition frequency of the current candidate node to any node in the initial evolutionary domain is higher than the preset evolutionary transition threshold, the current candidate node is added to the initial evolutionary domain. Otherwise, the reverse traversal expansion of the current candidate node is stopped. When there are no candidate nodes that can be expanded by reverse traversal, the traversal is stopped, and the target evolutionary domain after expanding the initial evolutionary domain is obtained.

5. The welding control optimization method for welded pipe manufacturing according to claim 4, characterized in that, Based on the target evolution domain, each state regression anchor point deviating from the origin is extracted, including: Based on the target evolution domain and the state transition path to which the candidate matching node belongs, identify the local state anchor point corresponding to each candidate matching node; Determine the multiple production stages of welded pipe production and identify the production stage to which multiple path nodes in each state transition path belong; Based on the production stage to which the path node belongs, determine the candidate stage range for each deviation from the origin. Based on the candidate stage range, select multiple stage matching nodes from the multiple candidate matching nodes contained in the deviation pattern node to which the deviation from the origin belongs. Extract the state signature difference and sequence position difference between the deviation from the origin and the local state anchor point of each stage matching node. Calculate the evolution matching parameters corresponding to the deviation from the origin and multiple local state anchor points based on the state signature difference and sequence position difference. Select the state regression anchor point corresponding to the deviation from the origin from multiple local state anchor points based on the evolution matching parameters.

6. The welding control optimization method for welded pipe manufacturing according to claim 5, characterized in that, Anomalies in the production process of the target welded pipe are monitored using state evolution maps. After identifying deviations in production, welding control optimization is performed based on an anomaly repair database, including: Welding monitoring data of the target welded pipe is collected, and real-time state signatures corresponding to the welding monitoring data in multiple state sampling segments are extracted to construct the real-time state transition path of the target welded pipe. Based on the state evolution graph, the real-time deviation point of the real-time state transition path is identified. The real-time deviation point is matched with multiple deviation pattern nodes to determine the target deviation pattern node corresponding to the real-time deviation point. Based on the anomaly repair database, the state regression anchor points corresponding to the target deviation pattern nodes in multiple production stages are determined. The optimization anchor point is determined according to the production stage to which the real-time deviation point belongs. The target optimization strategy is generated based on the optimization anchor point. The welding control of the target welded pipe is optimized based on the target optimization strategy.

7. A welding control optimization system for welded pipe manufacturing, characterized in that, The system is used to implement the welding control optimization method for welded pipe manufacturing as described in any one of claims 1-6, including: The welding data acquisition module is used to collect welding process record data of multiple welded pipes. The welding process record data includes welding time sequence data of multiple welding parameters and welding quality data of welded pipes, and identifies multiple state sampling segments of each group of welding time sequence data. The state transition analysis module is used to identify the parameter association state between any two welding parameters in each state sampling segment, construct the state signature of each state sampling segment with respect to multiple welding parameters, and construct the state transition path of each group of welding process recorded data. The sample deviation identification module is used to divide multiple sets of welding process record data into normal samples and abnormal samples based on welding quality data, construct a state evolution map based on the state transition paths of multiple normal samples, and identify the deviation point from the origin of each abnormal sample based on the state evolution map. The matching node generation module is used to construct multiple deviation pattern nodes based on multiple deviations from the origin, and to determine multiple candidate matching nodes for each deviation pattern node based on multiple normal samples. The state regression analysis module is used to determine multiple closed nodes in the state evolution map based on multiple normal samples and construct the target evolution domain. It also extracts the state regression anchor point for each deviation from the origin based on the target evolution domain. The welding control optimization module is used to build an anomaly repair database based on multiple state regression anchor points, monitor the production process of the target welded pipe through the state evolution map, and optimize welding control based on the anomaly repair database after identifying the production deviation state.

Images