A main transformation site wire digitization process modeling method and system

Abstract

The application relates to the technical field of production process optimization, and particularly relates to a main transformer substation conductor digital process flow modeling method and system. The method comprises the following steps: acquiring process parameter data in a whole conductor production process to form a complete data set, analyzing process parameter correlation modes to obtain parameter correlation strength indexes; judging process parameter linkage relationships according to the parameter correlation strength indexes and a preset threshold value, determining linkage parameter groups, and combining historical defect occurrence data and parameter change trends to construct a defect correlation data set; constructing a defect prediction model based on the defect correlation data set and determining defect positioning rules, combining real-time process parameter data to extract real-time linkage data groups, and realizing defect positioning; updating process parameter records according to positioning results, and determining a process adjustment path. The application solves the problems of difficult identification of cross-process parameter correlation, difficult prediction and positioning of defects and process optimization lag in the conductor production of the main transformer substation, and realizes accurate defect tracing in the production process and process optimization.

Description

Technical Field

[0001] This application relates to the field of production process optimization technology, and in particular to a digital process flow modeling method and system for main transformation conductors. Background Technology

[0002] The main transformer conductor is a critical conductor in the manufacturing of power transformer windings. Its production process typically involves multiple continuous steps, including wire drawing, annealing, transposition, stranding, insulation coating, drying, and finished product inspection. Because this type of conductor has high requirements for dimensional accuracy, transposition pitch, conductor tension consistency, insulation thickness, and electrical performance stability, fluctuations in any step parameter can be transmitted along the production chain and amplified in subsequent processes, ultimately leading to quality problems such as structural deviations, uneven insulation, and abnormal electrical performance in the finished product. Therefore, effectively modeling and digitally analyzing the entire main transformer conductor production process has become a crucial technical challenge for improving product quality and production stability.

[0003] Current production management methods for main transformer conductors largely rely on single-process monitoring, manual experience judgment, or post-production quality inspection. They typically monitor only individual parameters such as tension, temperature, speed, pitch, and coating thickness independently, lacking a systematic analysis of the interrelationships between parameters across processes. Because conductors operate continuously on the production line, changes in parameters from preceding processes often have length, positional differences, and time lags in their transmission to subsequent processes. Analyzing data based solely on samples taken at the same moment or from a single point can easily lead to misjudgments that do not conform to actual process transmission patterns, making it difficult to accurately reveal the defect formation mechanism. Furthermore, there is often a lack of unified correlation between historical production records, online monitoring data, and defect inspection results. This makes it difficult to quickly trace the specific abnormal process and key influencing parameters after quality anomalies occur, and also hinders the development of targeted process adjustment plans.

[0004] Therefore, for the continuous, multi-process, and strongly coupled manufacturing scenario of main transformation conductors, there is an urgent need for a digital process flow modeling method and system that can integrate full-process process data, identify parameter linkage relationships, predict defect risks, locate the source of anomalies, and guide process optimization, so as to improve the traceability of the production process, the accuracy of defect location, and the efficiency of process optimization. Summary of the Invention

[0005] This application provides a digital process flow modeling method and system for main transformer conductors, which is used to solve the problems of difficulty in identifying the linkage relationship of cross-process parameters, difficulty in tracing the cause of defects, and lag in process optimization during the production of main transformer conductors, thereby improving the accuracy of defect location, traceability of the production process, and stability of product quality.

[0006] In a first aspect, this application provides a method for modeling the digital process flow of a main transform conductor, the method comprising: Step S1: Obtain process parameter data for the entire wire production process, organize them according to time sequence and process steps, and form a complete dataset after preprocessing; combine wire length segment division and time delay correction to analyze the correlation pattern between process parameters and obtain parameter correlation strength index. Step S2: Determine the linkage relationship between process parameters based on the comparison result between the parameter correlation strength index and the preset threshold, and determine the linkage parameter group; extract defect occurrence data related to the historical linkage parameter group from historical production records, analyze the parameter value change trend, and construct a defect correlation dataset; Step S3: Construct a defect prediction model for the defect association dataset and determine the defect location rules; collect real-time process parameter data, extract real-time linkage data groups, and obtain the location results based on the real-time linkage data groups using the defect prediction model and the defect location rules; Step S4: Update the process parameter records in the dataset according to the positioning results, compare the abnormal process parameters in the linkage parameter group corresponding to the positioning results with the normal process parameter values ​​in the most recent linkage parameter group, and determine the process adjustment path based on the comparison difference and the potential defect causes.

[0007] Secondly, the present invention also provides a digital process flow modeling system for main transformation conductors, used to implement the above-mentioned method, the system comprising: The data acquisition unit is used to acquire process parameter data in the entire wire production process, organize them according to time sequence and process links, and form a complete dataset after preprocessing; The correlation analysis unit is used to analyze the correlation patterns between process parameters based on the dataset, combined with conductor length segmentation and time delay correction, and to obtain parameter correlation strength index. The defect association unit is used to determine the linkage relationship between process parameters based on the comparison result of the parameter association strength index and the preset threshold, and to determine the linkage parameter group; extract defect occurrence data related to the historical linkage parameter group from historical production records, analyze the parameter value change trend, and construct a defect association dataset; The defect localization unit is used to construct a defect prediction model for the defect association dataset, determine defect localization rules, collect real-time process parameter data, extract real-time linkage data groups, and obtain localization results based on the real-time linkage data groups through the defect prediction model and the defect localization rules. The process optimization unit is used to update the process parameter records in the dataset according to the positioning results, compare the abnormal process parameters in the linkage parameter group corresponding to the positioning results with the normal process parameter values ​​in the most recent linkage parameter group, and determine the process adjustment path based on the comparison difference and the potential defect causes.

[0008] Thirdly, the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0009] Technical effects of the invention This invention unifies the previously scattered parameter data from processes such as wire drawing, annealing, transposition, stranding, insulation coating, drying, and finished product inspection into a single conductor segment analysis unit by collecting parameters from the entire process, dividing the length into segments, and correcting for time delays. This allows different process parameters to correspond according to actual material transfer relationships, avoiding false correlations caused by analyzing data only at the same moment. Furthermore, by mining association rules to obtain parameter association strength indices and combining them with threshold judgments to form a linked parameter group, the transfer relationships between parameters such as preceding tension, temperature, speed, transposition pitch, coating thickness, and electrical properties can be quantified, providing a reliable data foundation for defect analysis. Based on this, historical defect occurrence data is combined with the parameter change trends of the linked parameter group to construct a defect association dataset, establishing a stable correspondence between defect types, abnormal processes, key parameters, and their change paths. Subsequently, a defect prediction model and defect location rules are used to analyze the real-time linked data group, enabling early identification of insulation defects in the main transposition conductor production process. In addition to addressing quality risks such as uniformity, transposition pitch deviation, abnormal conductor tension, and electrical performance fluctuations, this solution can further pinpoint the process to which abnormal parameters belong and their propagation path, enabling a shift from defect discovery to tracing the root cause. Simultaneously, the location results are fed back to the process parameter record and compared with the most recent normal linkage parameter group to determine the adjustment direction and process adjustment path for the abnormal parameters. This technical solution, through the coordination between each step, transforms the main transposition conductor production process from relying on manual experience and post-production inspection to proactive prediction, precise tracing, and dynamic optimization based on full-process data-driven approaches, improving defect location accuracy, timeliness of process adjustments, and product quality stability. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of a digital process modeling method for a main transformation conductor in an embodiment; Figure 2 This is a schematic diagram illustrating the defect propagation path and location in the embodiment; Figure 3 This is a structural diagram of a digital process flow modeling system for a main transformation conductor in an embodiment. Detailed Implementation

[0012] This application provides a method and system for modeling the digital process flow of a master transform conductor. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below, such as... Figure 1 As shown in the figure, an embodiment of this application provides a method for modeling the digital process flow of a main transformation conductor, including: Step S1: Obtain process parameter data for the entire wire production process to form a complete dataset; analyze the correlation patterns between process parameters based on the dataset to obtain parameter correlation strength indices; In step S1, a complete dataset is formed, including: The process parameters of each stage of wire production are collected in real time by sensors and monitoring equipment, and the operating status and parameter values ​​of each stage are recorded. The collected process parameter data are organized according to time sequence and process stage to generate a complete dataset containing the parameter change trends of each stage. The complete dataset is preprocessed to remove outliers and missing values, and the processed complete dataset is stored in a structured format.

[0014] Specifically, because the production of main transformer conductors involves multiple stages, such as wire drawing, stranding, insulation coating, and finished product inspection, each process generates a large amount of parameter data. This parameter data directly reflects the operating status of the production equipment and potential influencing factors on product quality. ，Real-time data collection of process parameters at each stage of wire production using sensors and monitoring equipment records the operational status and parameter values ​​of each process. For example, in the wire drawing process, temperature, pressure, and speed sensors are used to monitor parameters such as the temperature of the wire drawing machine, tensile force, rotational speed of the stranding machine, and coating thickness during insulation coating. These parameter data are recorded with timestamps to ensure data timeliness and traceability, allowing for precise capture of the operational status and changes of each process during production. Next, the collected process parameter data is organized according to time sequence and process steps, generating a complete dataset containing the trend of parameter changes at each stage. Specifically, the data is categorized and stored according to production batch and time sequence, with parameter data for processes such as wire drawing, stranding, and insulation coating stored separately and arranged along a timeline. This approach helps analyze the interrelationships between process parameters and provides a necessary foundation for subsequent parameter correlation analysis.

[0015] Then, the complete dataset is preprocessed to remove outliers and missing values. In actual production, the collected data may be affected by sensor malfunctions or human error, resulting in outliers or missing values, which in turn affects the accuracy of subsequent analysis. Therefore, the dataset must be cleaned. Specifically, by setting reasonable parameter ranges, data that clearly exceeds the normal range can be marked and removed. For example, when the tensile force value in the wire drawing process suddenly increases to a value greater than the outlier threshold, it is obviously outlier data and needs to be removed. The outlier threshold is determined by statistically analyzing the maximum tensile force value in the wire drawing process under normal historical conditions. For missing values, interpolation methods such as linear interpolation can be used to fill in the missing values, ensuring the continuity of the data over time. The preprocessed dataset will more realistically reflect the parameter changes in the production process, improving the accuracy of subsequent analysis. Finally, the processed complete dataset is stored in a structured format for subsequent analysis and to explore the potential relationships between process parameters. Specifically, after cleaning and preprocessing, the data will be stored in a structured format, such as a table, with each row representing a data record at a specific time point and each column corresponding to a process parameter such as tensile force or rotational speed. Structured storage facilitates the extraction of correlation patterns between parameters using data mining tools. For example, each process parameter stored in the database is assigned a unique identifier, enabling rapid location of specific process data records during subsequent analysis. This allows for further parameter correlation analysis, providing strong support for subsequent defect localization, process optimization, and production efficiency improvement. The above technical solutions ensure data quality and integrity during wire production, providing accurate data support for subsequent defect localization.

[0016] Further, in step S1, the parameter correlation strength index is obtained, including: The complete dataset is processed using an association rule mining algorithm to extract frequent itemsets between parameters of each process step; the support and confidence between process parameters are calculated based on the frequent itemsets to generate association rules between parameters; the influence between parameters is quantified through the association rules to obtain parameter association strength indicators; and the parameter association strength indicators are ranked.

[0017] Specifically, since the production of main-transformer conductors is a continuous, multi-stage manufacturing process, changes in the process status of preceding stages typically do not immediately affect subsequent stages. Instead, as the conductor material is transferred along the production line, these changes manifest as parameter variations or quality anomalies in subsequent stages only after a certain equipment interval and time delay. Therefore, to avoid false correlations caused by relying solely on the co-occurrence of data at the same moment, the length measuring device in the conductor production process is used as the segmentation basis. The continuously produced conductors are divided into multiple continuous length segments, with each length segment serving as an analysis unit. For example, every preset length of conductor is defined as an independent analysis unit. Compared to using only a fixed time window, using conductor length segments as analysis units more accurately reflects the transfer process of the same material across different stages.

[0018] Since the same conductor length segment appears at different times in different processes, time delay correction is performed on the parameter data of each process. Specifically, based on the physical distance between adjacent processes on the production line and the conductor running speed, the time delay required for a certain length segment to be transferred from the preceding process to the subsequent process is calculated. Let the physical distance between process i and process j be . The conductor's running speed is Then the transfer time delay from process i to process j Can be expressed as physical distance With the speed of the conductor The ratio, during data alignment, will synchronize the preceding processes at specific times. The collected parameter data is compared with the subsequent processes at different times. The corresponding parameter data are matched so that the parameter states of the same conductor length segment in different processes can be mapped to the same analysis unit. If the conductor speed fluctuates during production, the dynamic time delay can be calculated using the average speed of segments or the instantaneous speed integral method to improve the matching accuracy, thereby ensuring that the subsequent analysis reflects the actual process state changes of the same conductor segment in different processes.

[0019] After data alignment, the continuous process parameters in the complete dataset are discretized to construct state-based data suitable for association rule mining. These process parameters include at least the tensile force, temperature, and linear speed in the wire drawing process; the rotational speed, tension, and pitch deviation in the stranding process; the coating thickness, heating temperature, and coating pressure in the insulation coating process; and the dimensional deviations and electrical performance deviations in the finished product inspection process. For parameters with clear process target values ​​and allowable deviation ranges, state classification is performed according to the target value and allowable deviation. For example, let the average value of this parameter in historical qualified batches be... The standard deviation is Then it can be less than The value is recorded as a low state, and the position is... The value within is considered the normal state, and will be greater than The value is recorded as an elevated state, where The preset coefficient is preferably a value between 1 and 2. For volatility parameters, instead of using instantaneous sampled values ​​directly, the volatility characteristic values, such as standard deviation, range, or coefficient of variation, are calculated within each length segment or analysis window. Then, the parameters are classified into stable, slightly fluctuating, or violently fluctuating states according to preset thresholds. This avoids distortion of state judgment due to accidental deviations of a single sampling point.

[0020] Each conductor length segment is combined with multiple process parameter states to form a transaction record, which is used for association rule mining. Each transaction record includes at least a length segment identifier, a production batch identifier, and multiple process parameter state items corresponding to that length segment. For example, a transaction record corresponding to a certain conductor length segment can be represented as: Batch A, nth length segment, drawing tension is high, drawing temperature is normal, stranding rotation speed is fluctuating, insulation layer thickness is too thin, and finished product size deviation is too large. This transaction record does not represent the original set of sampled values ​​at the same moment, but rather the combination of states of the same conductor length segment after process transmission and time delay correction in different processes. Therefore, this transaction record can accurately reflect the actual linkage relationship of cross-process parameters in the production of the main transformer conductor. After forming the transaction dataset, an association rule mining algorithm is used to extract frequent itemsets between process parameters. The association rule mining algorithm can be the Apriori algorithm or the FP-Growth algorithm. When the data scale is large and the sampling frequency is high, the FP-Growth algorithm is preferred to reduce the candidate itemset generation process and improve the computational efficiency. Specifically, a minimum support threshold is first set, and the frequency of occurrence of each parameter state item and its combination in all transaction records is counted. When the occurrence ratio of a certain parameter state combination reaches or exceeds the minimum support threshold, the combination is identified as a frequent itemset. The minimum support threshold can be set according to the required analytical sensitivity.

[0021] After extracting frequent itemsets, association rules between parameters are generated based on these frequent itemsets. To enhance the process interpretability of the rules, in this embodiment, the antecedent of the rule is limited to the parameter state of the preceding process or the control parameter state in the same process, and the consequent of the rule is limited to the parameter state of the subsequent process or the quality result parameter state. Preferably, the consequent of the rule is a single parameter state item to facilitate subsequent defect location analysis. For example, association rules of the following form can be generated: high wire drawing tension corresponds to fluctuating stranding rotation speed; fluctuating stranding rotation speed corresponds to thin insulation layer thickness; thin insulation layer thickness corresponds to out-of-tolerance finished product inspection dimensional deviation. For each association rule, support, confidence, and lift are calculated respectively. Support is used to represent the frequency of simultaneous occurrence of the antecedent and consequent of the rule in all transaction records, and the calculation formula is: in, This indicates that the antecedent of the rule is satisfied simultaneously. and rule consequent The number of transaction records, The total number of transaction records is represented; the confidence score represents the probability that the consequent of a rule will occur given that the antecedent of the rule has occurred, and is calculated using the following formula: Lift measure whether the occurrence of the antecedent of a rule increases the likelihood of the consequent occurring. The formula is as follows: in Lift indicates the support of the rule consequent Y in all transaction records, which is the proportion of transaction records containing Y to the total number of transaction records. When the lift is greater than 1, it indicates that the occurrence of the rule antecedent has a positive reinforcing effect on the rule consequent; when the lift is close to 1, it indicates that the two are more likely to have only a general co-occurrence relationship.

[0022] Since the generation of frequent itemsets and association rules is based on transaction co-occurrence statistics, reverse rules that do not conform to the actual causal path of production may appear. Therefore, the generated association rules are further screened based on process timing constraints. Specifically, only association rules that meet the following conditions are retained: the process to which the parameter corresponding to the rule's predecessor belongs is no later than the process to which the rule's successor belongs; when the rule's predecessor and successor come from different processes, the rule's predecessor is preferably from the preceding process of the rule's successor; when the rule's predecessor and successor come from the same process, the rule's predecessor should be a control parameter or a state parameter, and the rule's successor should be a result parameter or a quality evaluation parameter. Association rules that do not meet the above conditions, such as rules that use the subsequent finished product inspection results to inversely indicate the tensile force of the preceding wire drawing process, are considered pseudo-associations that do not conform to the process transmission logic and are eliminated. This screening method ensures that the retained association rules better reflect the actual parameter transfer relationships in the production of the main transform conductor. After obtaining association rules that meet the process timing requirements, each association rule is comprehensively quantified to obtain a parameter association strength index. This index considers not only statistical support and confidence but also process distance and quality correlation to more accurately reflect the actual influence between different parameters. The calculation formula is given below. In one embodiment, process distance weights are first defined. Let the number of process intervals between the process to which the preceding item belongs and the process to which the following item belongs be . The process distance weight can be calculated using the following formula: The weight of process distance gradually decreases, thus reflecting the attenuation characteristics of cross-process influence transmission; further defining quality-related weights. The aforementioned quality-related weights do not depend on defect labels in subsequent steps, but are preset based on the category of the rule's consequent parameter. Specifically, when the rule consequent is a final product inspection parameter, it is assigned the first weight value; when the rule consequent is a key process parameter directly affecting product quality, it is assigned the second weight value; and when the rule consequent is a general process parameter, it is assigned the third weight value. The first weight value is greater than the second weight value, and the second weight value is greater than the third weight value. This setting ensures that parameters directly related to the quality result are given priority in subsequent sorting.

[0023] In a preferred embodiment, the parameter correlation strength index Calculate using the following formula: in, Indicates support level, Indicates the confidence level. Indicates the distance weight between processes. Indicates quality-related weights, This indicates the lift after normalization. , , , Let be the weighting coefficients, and satisfy: ,in, This is the original lift, with values ​​ranging from 0 to 1. , , , This can be determined through historical sample verification. Specifically, multiple sets of candidate weight coefficients are used to calculate the parameter correlation strength index, and based on the comprehensive evaluation results of its coverage of historical quality anomaly samples, false alarm rate, and rule interpretability, the set of coefficients with the best score is selected as the practical application value and as a preferred example. , , , This allows each association rule to be mapped to a parameter association strength index within a uniform numerical range, facilitating horizontal comparisons between different rules. Furthermore, the association rules are sorted from largest to smallest according to their corresponding parameter association strength indices, laying the foundation for subsequent judgments on linkage relationships.

[0024] Step S2: Determine the linkage relationship between process parameters based on the parameter correlation strength index, and determine the linkage parameter group; extract relevant data from the linkage parameter group to construct a defect correlation dataset; In step S2, determining the linkage parameter group includes: A preset threshold is set as the judgment standard, and the magnitude of the parameter correlation strength index is compared with the preset threshold. If the parameter correlation strength index exceeds the preset threshold, it is determined that there is a linkage relationship between the corresponding process parameters. The process parameters with the linkage relationship are grouped into linkage parameter groups. The linkage parameter groups are classified and archived, and the correlation mode and strength value of the parameters in each group are recorded.

[0025] Specifically, based on the mean and fluctuation range of parameter correlation strength indicators under historical normal production conditions, a preset threshold that can distinguish between normal random fluctuations and stable linkage relationships is determined. To reflect the actual operating status of the current production line while avoiding threshold deviations due to individual abnormal records, the preset threshold is not applied to all process parameters with identical fixed values. Instead, it is set in layers based on the process type to which the parameter belongs and the quality correlation of subsequent parameters. Specifically, for subsequent parameters that are finished product quality inspection parameters, key process quality parameters, or parameters directly related to defects in the main transformer conductor, a relatively low preset threshold can be used to avoid missing low-frequency but high-risk linkage relationships. For general equipment operating status parameters, a relatively high preset threshold can be used to reduce false correlations caused by ordinary fluctuations. The aforementioned key process quality parameters may include transposition pitch deviation, conductor tension abnormalities, insulation layer thickness deviations, coating temperature abnormalities, etc. Through this layered threshold setting method, both the sensitivity of defect prevention and the accuracy of linkage relationship judgment can be considered.

[0026] After setting the preset threshold, the parameter correlation strength index in each parameter correlation record is compared with the corresponding preset threshold. When the parameter correlation strength index of a parameter correlation record is greater than or equal to its corresponding preset threshold, it is determined that there is a linkage relationship between the process parameters corresponding to that parameter correlation record; when the parameter correlation strength index is less than the corresponding preset threshold, it is determined that there is no effective linkage relationship between the process parameters corresponding to that parameter correlation record. To avoid misjudgment due to single sampling anomalies or short-term disturbances, process transfer direction constraints and stability verification conditions can also be introduced during the comparison process. That is, only when the same parameter correlation relationship satisfies the above process transfer... A linkage is considered valid only when the threshold condition is met in multiple consecutive adjacent conductor length segments. If the threshold condition is met only in a single window, it can be temporarily marked as a linkage to be observed and re-evaluated after subsequent data updates. The aforementioned process transfer direction constraint means that the direction of the linkage should be consistent with the production process of the main transform conductor, that is, the process to which the preceding parameter belongs should not be later than the process to which the following parameter belongs, or in the same process, the preceding parameter should be a control parameter and the following parameter should be a status parameter or a quality evaluation parameter. By combining strength comparison, direction constraint and stability verification, the erroneous linkage judgment caused by statistical randomness can be effectively reduced.

[0027] After confirming the parameters with interrelationships, the process parameters with interrelationships are combined into a linkage parameter group. This linkage parameter group can be a binary parameter group consisting of two parameters or a multi-parameter parameter group consisting of multiple parameters. For binary parameter association records, the preceding and following parameters can be directly combined into a linkage parameter group. For multiple association records with common parameters or forming a continuous process transfer chain, they can be further merged to form a multi-parameter linkage parameter group. For example, if there is a linkage between abnormal wire drawing tension and stranding rotation speed fluctuations, and also a linkage between stranding rotation speed fluctuations and insulation layer thickness deviations, then the wire drawing tension, stranding rotation speed, and insulation layer thickness can be combined into a cross-process linkage parameter group. This linkage parameter group can characterize the parameter transfer path from the preceding wire drawing to the subsequent coating, which is more conducive to subsequent identification of the causes of defects. When combining multiple linkage parameters, a parameter association graph is preferred. Specifically, each process parameter is treated as a node, and parameter pairs with confirmed linkage relationships are treated as directed edges. The direction of these directed edges aligns with the process transfer direction, and the edge weights are the corresponding parameter association strength indices. Subsequently, connected paths or strongly connected structures are searched within the parameter association graph, and the set of nodes that meet the process continuity and strength requirements is determined as the linkage parameter group. For paths spanning multiple processes, the association strength between adjacent nodes in the path can be required to meet the aforementioned preset thresholds. For cases with multiple branch paths, the main linkage path can be selected based on the edge with the maximum association strength or its proximity to quality parameters. This expands isolated binary parameter relationships into parameter linkage groups that conform to the production process chain, thereby more accurately reflecting the multi-factor coupling relationships in the formation process of defects in the main transformer conductor.

[0028] Subsequently, the identified linkage parameter groups were categorized and archived according to dimensions such as process link type, parameter category, defect relevance direction, and linkage strength level. Specifically, according to process link type, they can be divided into single-process internal linkage parameter groups, adjacent process linkage parameter groups, and multi-process linkage parameter groups; according to parameter category, they can be divided into equipment operation, process control, material status, and quality inspection linkage parameter groups; according to defect relevance direction, they can be divided into linkage parameter groups related to transposition pitch deviation, conductor tension abnormalities, insulation layer thickness abnormalities, and finished product electrical performance abnormalities. For each linkage parameter group, the archived information includes at least the group number, parameter names within the group, the process to which it belongs, parameter status, correlation direction, correlation mode, correlation strength index between each parameter, and the analysis batch and time range on which the group was based. Among them, the correlation mode is used to describe the linkage form between parameters within the group. For a binary linkage parameter group, the correlation pattern can be represented as an increase in the probability of a change in the state of parameter B due to a change in the state of parameter A. For a multi-element linkage parameter group, the correlation pattern can be represented as a chain pattern of abnormal preceding parameters, fluctuations in intermediate process parameters, and deviations in subsequent quality parameters. It can also be represented as a convergence pattern in which multiple preceding parameters jointly affect a single subsequent parameter, or a dispersion pattern in which a single preceding parameter simultaneously affects multiple subsequent parameters. By recording the correlation pattern, subsequent defect analysis can not only identify which parameters are related, but also the transmission direction and combination structure between parameters.

[0029] Further, in step S2, a defect association dataset is constructed, including: Extract defect occurrence data related to the linkage parameter group from historical production records; analyze the specific parameter value change trend of each linkage parameter group in the defect occurrence data; map the parameter value change trend to the defect type to generate a defect association dataset containing defect features and parameter change relationships; clean and format the defect association dataset.

[0030] Specifically, defect occurrence data related to the aforementioned linkage parameter group is extracted from historical production records. These historical production records preferably include production batch information, work order information, product specification information, process parameter records, conductor length measurement records, online inspection records, offline quality inspection records, defect registration records, and rework or scrap records. The defect occurrence data includes at least the defect type, defect occurrence process, defect discovery time, corresponding conductor segment location, defect severity, and batch identifier. For defects discovered through online inspection, such as abnormal insulation thickness, transposition pitch deviation, structural inhomogeneity caused by conductor tension fluctuations, and surface damage, defect markings can be directly established based on the inspection time and length location. For defects identified only during the finished product inspection stage, such as unstable finished product electrical performance or inconsistent local insulation, the process parameter data from each process stage is traced back based on the conductor length segment, batch number, reel number, or encoder length information recorded during inspection.

[0031] When extracting defect occurrence data, the conductor length segment is used as the basic matching unit. Specifically, each linkage parameter group determined in step S2 is mapped to the same conductor length segment in the main transformation conductor production process, and a correspondence between the length segment and the defect record is established. For cases where there are multiple defect tags for the same length segment, a multi-tag recording method can be used for storage. After completing the defect sample extraction, the specific parameter value change trend of each linkage parameter group in the above defect occurrence data is analyzed, that is, the change trajectory, fluctuation characteristics and mutual transmission relationship of each parameter in the linkage parameter group within the preset range before and after the defect formation. Specifically, for each defect record, the original process data of each parameter in the linkage parameter group is extracted forward and backward, centered on the conductor length segment or defect occurrence time, to form a defect correlation analysis window. The scope of the window can be set according to the material transfer distance between processes, the running speed of the conductor, and the parameter response lag characteristics, so that the window includes both the induction stage data before the defect occurs and the defect manifestation stage data. Subsequently, the trend of each parameter in the analysis window is extracted. Preferably, at least one or more of the following characteristics are calculated for each parameter in the linkage parameter group: the offset from the set value, the rate of change, the proportion of the parameter staying outside the normal range, and the order of change between parameters. For parameters that can reflect stability, the variance or standard deviation can also be calculated. For equipment parameters with periodic fluctuation characteristics, the number of periodic anomalies or the change in fluctuation frequency can also be calculated. Through the above calculations, each linkage parameter group can form a structured trend description result for each defect sample, rather than just isolated instantaneous measurement values.

[0032] For multi-parameter linkage relationships, it is also necessary to analyze the coordinated change trend among parameters within the group. Specifically, based on the correlation direction determined in step S2, it is necessary to determine whether the anomaly of the preceding parameter occurs before the anomaly of the subsequent parameter, and whether the anomaly propagates gradually along the process flow. For example, in a certain linkage parameter group, if the wire drawing tension deviates from the set value, the stranding rotation speed fluctuates continuously, and subsequently the insulation layer thickness deviates on the corresponding conductor segment, then this sample can be marked as a typical chain-like transmission trend. To achieve this analysis, the abnormal starting point position of each parameter within the group can be identified, and the sequence and interval distance between each abnormal starting point can be compared. If the sequence is consistent with the process direction and the interval is within a reasonable transmission range, then the linkage parameter group is considered to have formed an effective transmission trend in the defect sample. If multiple parameters fluctuate simultaneously but have no obvious sequential relationship, it can be marked as a common-cause disturbance type change trend. By distinguishing between the above-mentioned chain-like transmission trend, effective transmission trend, and common-cause disturbance trend, the ability of the defect association dataset to express the actual defect mechanism can be improved.

[0033] After completing the parameter value change trend analysis, the above parameter value change trends are mapped to defect types to generate a defect association dataset containing defect features and parameter change relationships. Specifically, each defect sample or each defect conductor segment is taken as a data unit. The defect type label, defect severity label, defect process label, and trend features of the corresponding linkage parameter group corresponding to the data unit are combined and encoded. The above trend features can include single-parameter trend features and multi-parameter relationship features. Among them, single-parameter trend features are used to characterize the offset, fluctuation, and rate of change of each parameter itself, while multi-parameter relationship features are used to characterize the sequence, synchronization, transmission link, and linkage strength matching of parameter anomalies within the group, i.e., the chain transmission trend, effective transmission trend, and common cause disturbance type change trend. For the case where the same defect type corresponds to multiple linkage parameter groups, it can be stored by using one defect sample to correspond to multiple parameter group features; or it can be merged around the defect type to form a hierarchical structure data of defect type, candidate linkage parameter groups, and trend features.

[0034] In terms of data organization, the defect association dataset is preferably constructed in a data format that combines tabular and graph structures. Each row corresponds to a defect sample or a conductor segment sample, and each column corresponds to a trend feature field and a defect label field for a linkage parameter group. The graph structure data is used to preserve the transmission relationship within the linkage parameter group, where nodes represent parameter state features, edges represent directional linkage relationships between parameters, and edge attributes record the trend strength or propagation order under the defect sample. This dual representation method allows subsequent model training to utilize both structured features and, when needed, process relationship topology information, improving the interpretability of defect location results. After generating the defect association dataset, the dataset is cleaned and formatted. Data that clearly does not conform to process rules is subject to anomaly verification, such as parameter mutations without equipment event records, inconsistencies between defect labels and length segment positions, and reversed time order of adjacent process data. These are preferably marked as suspicious samples and manually verified. Furthermore, data fields from different sources are uniformly formatted, including uniform time format, length segment number format, parameter name encoding, defect type encoding, and process identifier encoding. For trend features with different dimensions, normalization or standardization can be performed to avoid certain large numerical dimensions from having an excessive impact on the results during subsequent modeling.

[0035] Through the above technical solution, based on the linkage parameter group already selected in step S2, real samples related to defects are further extracted from historical production records. The changing trends and transmission patterns within the parameter group before and after the formation of defects are analyzed, and the above trend characteristics are stably correlated with defect types to form a defect association dataset with clear structure, unified fields, and clear process meaning. This dataset not only retains the process characteristics of multi-process, multi-parameter, and continuous transmission in the production of the main transformation conductor, but also provides a data foundation that can be directly used in the future.

[0036] Step S3: Construct a prediction model for the defect association dataset and determine the defect location rules; process real-time data through the defect location rules, obtain abnormal parameter values ​​and determine the cause of the defect to obtain the location result; In step S3, the defect location rules are determined, including: A defect prediction model is constructed by training the defect association dataset using the random forest algorithm. The parameter value characteristics of each historical linkage parameter group when the defect occurs are analyzed through the defect prediction model to extract potential defect causes. Based on the correspondence between the potential defect causes and the historical linkage parameter groups, defect location rules are generated. The defect location rules are verified and the model parameters are adjusted.

[0037] Specifically, in order to obtain accurate defect location, the random forest algorithm is used to train the above-mentioned defect association dataset to build a defect prediction model. This algorithm can handle multi-dimensional features, nonlinear relationships and multivariate coupling scenarios, and has good adaptability to the common situation in the production of main transformation conductors where multiple process parameters work together and the boundaries of different defect types are not completely linearly separable.

[0038] During implementation, the aforementioned defect association dataset is prepared for modeling. Preferably, each sample in the defect association dataset is defined as an analysis unit corresponding to a conductor length segment. If there is a single defect or a compound defect, it is assigned a corresponding label, including at least a defect type classification. Analysis units under historical normal conditions are also marked as normal samples. Before model training, the samples are divided into training, validation, and test sets. Considering that the process parameters of adjacent conductor segments within the same batch are highly correlated, simple random division may easily result in samples from similar sources appearing in the training and test sets, leading to overly optimistic evaluation results. Therefore, it is preferable to group the samples according to production batch, reel number, product specification, or time period, that is, to ensure that adjacent conductor segments in the same batch appear in the same subset as much as possible.

[0039] After data partitioning, a defect prediction model is established using the aforementioned random forest algorithm. Specifically, the random forest consists of multiple decision trees. Each decision tree generates a subset of samples from the training samples through sampling with replacement, and then randomly selects some features from all features to participate in node partitioning. During node partitioning, for the classification task, the contribution of different candidate features and their candidate split points to the improvement of sample purity is calculated. The features and splitting conditions that enable the current node to distinguish defect categories are selected as the basis for partitioning. After continuous iteration, multiple decision trees are formed. Finally, the prediction results are output through majority voting or average class probability. The prediction results include the defect type and its corresponding probability.

[0040] After completing model training, the parameter value characteristics of each historical linkage parameter group at the time of defect occurrence are analyzed using the above prediction model to extract potential defect causes. The aforementioned potential defect causes are not simply regarded as high-importance parameters directly as causes, but rather are comprehensively identified by combining the parameter change trends within the linkage parameter group, process sequence, defect type, and the contribution of the group's features in the model to the prediction results. Specifically, potential defect causes are extracted from both global and local levels. At the global level, the importance and weight of all features in the trained random forest model, i.e., the above defect prediction model, are statistically analyzed. For each linkage parameter group, the importance of all features within the group is weighted to obtain an importance score at the historical linkage parameter group level. If the aggregate importance of a certain linkage parameter group in a specific defect type model is consistently higher than that of other parameter groups, it indicates that there is a stable correlation between the linkage parameter group and this type of defect. Furthermore, the features within the group are sorted by importance to identify the most critical trend features, such as persistently high preceding tension, amplified speed fluctuations in intermediate processes, and decreased insulation thickness in subsequent processes. At the local level, for each actual defect sample, it is input into the prediction model, and the classification path of each decision tree for that sample is recorded. The node splitting conditions that frequently appear in most voting trees are statistically analyzed. For example, in a large number of samples predicted as abnormal insulation thickness, if multiple trees contain conditions such as stranding tension fluctuation exceeding a certain state boundary, coating temperature being abnormally low, and the sequence of abnormalities in historical linkage parameter groups satisfying the path conditions of propagation from the preceding to the following sequence, then the combination of the above conditions can be identified as constituting the potential causal characteristics of this type of defect. In the above process, a single instantaneous parameter value is not directly used as the cause. Instead, the causes of single-parameter anomalies are extracted, namely, the continuous offset, abnormal fluctuation or rapid change of a certain parameter; the causes of chain transmission, namely, after the preceding parameter in the linkage parameter group is abnormal, the intermediate process and the subsequent quality parameters are abnormal in sequence; and the causes of composite coupling, namely, after multiple preceding parameters change together, the quality parameter deviation of the same subsequent part is caused. This can avoid the fact that some defects do not directly manifest as a single parameter exceeding the limit, but are reflected in the slight mismatch between transposition pitch, conductor torsion state and subsequent insulation coating consistency.

[0041] Defect localization rules are generated based on the correspondence between potential defect causes and historical linkage parameter groups. Specifically, for each defect type, historical linkage parameter groups and their key feature patterns that frequently appear and are of high importance in the model are summarized. For example, the following logical structure can be formed: a certain defect type corresponds to linkage parameter group A and linkage parameter group B, where A is mainly characterized by high and persistent preceding tension, and B is mainly characterized by increased subsequent temperature fluctuations. For each potential cause, the splitting conditions that repeatedly appear in the random forest decision path are extracted, and these conditions are transformed into executable rule triggering conditions. To avoid overly complex rules, features that appear frequently and have clear process significance within the same linkage parameter group are preferred as primary conditions, and other auxiliary conditions are added as supplementary conditions. For example, if the preceding traction tension in the linkage parameter group is abnormally high, and the rotation speed fluctuation in the intermediate process continues, while the subsequent insulation thickness shows a continuous downward trend, then the defect is determined to be preferentially located in the preceding traction and stranding collaborative control link. Each defect localization rule preferably includes at least: corresponding defect type, triggering condition, suspected problematic process, suspected core parameter, linkage parameter group identifier, process propagation path, and confidence level. The confidence level can be determined by combining the hit rate, coverage and false alarm rate of the rule in the validation samples.

[0042] Since multiple trees in a random forest may produce similar but slightly different combinations of conditions, it is preferable to merge the rules. Rules that point to the same defect type, the same problematic process, and have similar key triggering conditions are merged, retaining their common conditions and using the differences as supplementary explanations. This avoids an excessive number of rules that could affect field applications. In a preferred embodiment, the defect location rule not only outputs which type of parameter is abnormal, but also outputs the process chain that should be traced first. For example, for a detected problem of uneven insulation in finished products, the rule can indicate to prioritize checking the stranding tension, transposition pitch, and coating thickness. This chain of linkages, rather than simply prompting to check the final coating process, reflects the causal tracing process from the preceding cause to the subsequent manifest defect.

[0043] After generating defect localization rules, these rules need to be validated, and model parameters adjusted to improve prediction accuracy and ensure the applicability and reliability of the rules. Validation can be carried out at two levels: model validation and rule validation. Specifically, test set samples are input into the trained random forest model, and the accuracy, recall, precision, F1 score, and confusion of each defect type are statistically analyzed. For defect localization scenarios, special attention should be paid to the recall rate of defect samples, as missed detections will directly affect the subsequent localization effect. If the model has a high overall accuracy but a low recall rate for a few key defect categories, the class weights, sample balancing strategies, or feature selection methods should be adjusted first. The generated defect localization rules are also applied to independent validation samples one by one to check whether the rules can correctly point to the actual abnormal process and parameter group. Specifically, the number of rule hits, the proportion of correctly located processes after a hit, the proportion of defect samples covered by the rule, and the proportion of false triggers can be statistically analyzed. If a rule predicts the defect type relatively accurately, but on-site verification shows that the localization process is too far back or too far forward, it is necessary to trace back the potential cause extraction process, adjust the weight of the linkage parameter group, or reduce the rule triggering conditions.

[0044] After the model and rules have been validated, the defect location rules are compiled into a rule library. Each rule in the rule library preferably includes the following fields: rule number, applicable product specifications, applicable production line type, target defect type, trigger condition description, corresponding linkage parameter group, corresponding process, priority investigation parameters, rule generation basis version, and update time. When running new production batches, real-time parameter data can be directly mapped to the defined linkage parameter group feature space, and the rule library can be called for rapid defect location. Through the above technical solution, a defect prediction model is established using random forest, and potential defect causes are extracted from the model. Combined with the linkage parameter group, defect location rules are generated. This not only identifies the occurrence of defects but also locates the specific problematic process, parameter linkage chain, and abnormal mode in the main transformer conductor production process, thereby meeting the traceability requirements for defects in this specific application scenario.

[0045] Further, in step S3, the positioning result is obtained, including: Collect real-time process parameter data of the current conductor production; obtain real-time linkage data group based on the real-time process parameter data, input the real-time linkage data group into the defect prediction model, obtain defect prediction results, when the defect prediction result is abnormal, match the defect type and corresponding linkage data group in the defect prediction result with the defect location rules; identify abnormal parameter values ​​through the matching analysis, determine whether the abnormal parameter values ​​meet the characteristics of known potential defect causes; generate defect location results based on the determination results.

[0046] Specifically, after acquiring real-time process parameter data, alignment and preprocessing are performed, and real-time linkage data groups are extracted. This process is consistent with the feature extraction process when constructing the defect association dataset in step S2. The aforementioned linkage data groups are input into the aforementioned defect prediction model to obtain defect prediction results. For each conductor length segment, the model receives its corresponding real-time linkage data groups as input and outputs a defect prediction result. The aforementioned defect prediction result includes a predicted defect category label and a corresponding confidence level, i.e., anomaly probability. For example, for a certain length segment, the prediction result may be {prediction category: insulation layer thickness anomaly, confidence level: 0.85} or {prediction category: normal, confidence level: 0.98}. This prediction result provides a quick preliminary judgment on whether there is a defect risk in the current production state.

[0047] When the above defect prediction result is abnormal, the predicted defect type and corresponding linkage data group in the above defect prediction result are matched and analyzed with the above defect location rules to achieve targeted location. The system no longer blindly traverses and matches all rules, but only calls a subset of rules related to the predicted defect type. For example, if the defect prediction result is abnormal insulation layer thickness, only all rules with abnormal insulation layer thickness as the target defect type are selected from the rule base for matching, which significantly improves the efficiency and targeting of online analysis.

[0048] During the matching analysis, abnormal parameter values ​​are identified, and it is determined whether these abnormal parameter values ​​conform to the characteristics of known potential defect causes. For each selected defect location rule, its triggering condition is compared with the data from the linked data group that caused the abnormal prediction result in the previous stage. The triggering condition defines a typical parameter pattern for a potential defect cause, such as an abnormally high preceding traction tension, continuous fluctuations in the rotational speed of the intermediate process, and a continuous decreasing trend in the subsequent insulation thickness. The matching process checks whether the characteristic values ​​of the current linked data group meet these conditions. For example, for the sub-condition of an abnormally high preceding traction tension in the rule, it checks whether the traction tension-related characteristic values ​​in the corresponding linked data group fall within its defined abnormally high range. At the same time, it verifies whether the order of occurrence of the anomalies conforms to the process propagation chain defined in the rule to confirm whether the current parameter state truly conforms to the characteristics of known potential defect causes.

[0049] Finally, based on the above judgment results, defect location results are generated, recording the specific location of abnormal parameters and possible defect types. If no corresponding defect location rule is matched in any linkage parameter group, it indicates that the probability of a defect occurring is low. Conversely, if a linkage parameter group successfully matches one or more defect location rules, it indicates that the probability of a defect occurring is high, and a detailed defect location report is generated. This report is preferably in a structured record format, including at least: the current batch number, reel number, and product specifications; the start and end positions of the abnormal conductor segment, represented by uniform length coordinates, which can be simultaneously converted into radial layers and circumferential intervals on the reel for on-site search; the triggered defect location rule number; the suspected abnormal process directly pointed to by the rule; the linkage parameter group identifier that triggered the rule and the specific abnormal process parameters and abnormal values ​​within it; confirmation through the above rules is more interpretable than preliminary prediction; the above technical solution, through the strategy of prediction followed by location, effectively balances real-time performance and location depth, fully adapting to the actual needs of rapid defect response and accurate source tracing in the continuous production of main transformer conductors; Figure 2 As shown, the abnormally high tension in the wire drawing process is transmitted sequentially along the process chain to the stranding process, causing fluctuations in rotational speed. These speed fluctuations then affect the insulation coating process, resulting in uneven insulation thickness. Ultimately, this manifests as an electrical performance instability defect in the finished product inspection process. During the location phase, by triggering the corresponding defect location rules, it is possible to directly trace back to key linkage parameter groups such as stranding tension, transposition pitch, and coating thickness. This enables precise tracing from the finished product defect to the preceding abnormal process, demonstrating the ability to analyze the causal chain of cross-process defects.

[0050] Step S4: Update the dataset based on the positioning results and reanalyze the parameter correlation strength index to determine the process adjustment path; specifically including: The defect location results are fed back to the dataset to update the process parameter records in the dataset; the abnormal process parameters in the associated parameter group corresponding to the defect location results are compared with the normal process parameter values ​​in the same associated parameter group corresponding to the nearest analysis unit, and the adjustment direction of the abnormal process parameters is identified based on the comparison difference and potential causes; the final process adjustment path is generated through the adjustment direction to guide the optimization of the production process.

[0051] Specifically, for a confirmed defect location result, key information is extracted, namely the precise location of the faulty conductor segment, the triggered defect type (the aforementioned defect type), the located linkage parameter group, and the abnormal process parameter values ​​within the group. Subsequently, the original process parameter record corresponding to the aforementioned defect location is found in the dataset. The abnormal process parameter values ​​within the aforementioned linkage parameter group are compared with the normal process parameter values ​​corresponding to the linkage parameter group at the analysis unit position closest to the aforementioned defect location conductor segment. The comparison difference corresponding to each abnormal process parameter value is determined, and the corresponding abnormal parameter values ​​are adjusted in reverse according to the potential defect causes corresponding to the aforementioned defect. That is, the abnormal process parameters are adjusted sequentially according to the order of the abnormal process parameters in the corresponding process sequence or the order of occurrence of abnormal process parameters in the same process. The adjustment order and the comparison difference corresponding to the abnormal process parameters are used as the adjustment direction of the process parameters. Based on the adjustment direction, the final process adjustment path is generated. That is, according to the sequence of abnormal processes in the production process, combined with the difference between each abnormal parameter and its normal value, the specific parameter adjustment amount of each step is planned in reverse, thereby forming a quantitative and operable process adjustment path from the preceding to the following sequence. This guides the optimization of the aforementioned production process; through the cooperation of the above technical solutions, the abnormal process parameters can be accurately adjusted after obtaining the corresponding defect location results, thereby achieving the goal of process optimization and improving the quality of the aforementioned main transformation conductor.

[0052] This invention also provides a digital process flow modeling system for main transformation conductors, used to implement the above-mentioned method, such as... Figure 3 As shown, the system includes: The data acquisition unit is used to acquire process parameter data in the entire wire production process, organize them according to time sequence and process links, and form a complete dataset after preprocessing; The correlation analysis unit is used to analyze the correlation patterns between process parameters based on the dataset, combined with conductor length segmentation and time delay correction, and to obtain parameter correlation strength index. The defect association unit is used to determine the linkage relationship between process parameters based on the comparison result of the parameter association strength index and the preset threshold, and to determine the linkage parameter group; extract defect occurrence data related to the historical linkage parameter group from historical production records, analyze the parameter value change trend, and construct a defect association dataset; The defect localization unit is used to construct a defect prediction model for the defect association dataset, determine defect localization rules, collect real-time process parameter data, extract real-time linkage data groups, and obtain localization results based on the real-time linkage data groups through the defect prediction model and the defect localization rules. The process optimization unit is used to update the process parameter records in the dataset according to the positioning results, compare the abnormal process parameters in the linkage parameter group corresponding to the positioning results with the normal process parameter values ​​in the most recent linkage parameter group, and determine the process adjustment path based on the comparison difference and the potential defect causes.

[0053] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0054] In summary, this invention, through full-process parameter acquisition, length segment division, and time-delay correction, unifies the originally scattered parameter data from processes such as wire drawing, annealing, transposition, stranding, insulation coating, drying, and finished product inspection into a single conductor segment analysis unit. This allows different process parameters to correspond according to actual material transfer relationships, avoiding false correlations caused by analyzing data only at the same moment. Furthermore, by mining association rules to obtain parameter association strength indices and combining them with threshold judgments to form a linkage parameter group, the transfer relationships between parameters such as preceding tension, temperature, speed, transposition pitch, coating thickness, and electrical properties can be quantified, providing a reliable data foundation for defect analysis. Based on this, historical defect occurrence data is combined with the parameter change trends of the linkage parameter group to construct a defect association dataset, establishing a stable correspondence between defect types, abnormal processes, key parameters, and their change paths. Subsequently, a defect prediction model and defect location rules are used to analyze the real-time linkage data group, enabling early identification of insulation defects in the main transposition conductor production process. In addition to addressing quality risks such as uniformity, transposition pitch deviation, abnormal conductor tension, and electrical performance fluctuations, this solution can further pinpoint the process to which abnormal parameters belong and their propagation path, enabling a shift from defect discovery to tracing the root cause. Simultaneously, the location results are fed back to the process parameter record and compared with the most recent normal linkage parameter group to determine the adjustment direction and process adjustment path for the abnormal parameters. This technical solution, through the coordination between each step, transforms the main transposition conductor production process from relying on manual experience and post-production inspection to proactive prediction, precise tracing, and dynamic optimization based on full-process data-driven approaches, improving defect location accuracy, timeliness of process adjustments, and product quality stability.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0056] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0057] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for modeling the digital process flow of a main transform conductor, characterized in that, The method includes: Step S1: Obtain process parameter data for the entire wire production process, organize them according to time sequence and process steps, and form a complete dataset after preprocessing; combine wire length segment division and time delay correction to analyze the correlation pattern between process parameters and obtain parameter correlation strength index. Step S2: Determine the linkage relationship between process parameters based on the comparison result between the parameter correlation strength index and the preset threshold, and determine the linkage parameter group; extract defect occurrence data related to the historical linkage parameter group from historical production records, analyze the parameter value change trend, and construct a defect correlation dataset; Step S3: Construct a defect prediction model for the defect association dataset and determine the defect location rules; collect real-time process parameter data, extract real-time linkage data groups, and obtain the location results based on the real-time linkage data groups using the defect prediction model and the defect location rules; Step S4: Update the process parameter records in the dataset according to the positioning results, compare the abnormal process parameters in the linkage parameter group corresponding to the positioning results with the normal process parameter values ​​in the most recent linkage parameter group, and determine the process adjustment path based on the comparison difference and the potential defect causes.

2. The method according to claim 1, characterized in that, In step S1, a complete dataset is formed, including: The process parameters of each stage of wire production are collected in real time by sensors and monitoring equipment, and the operating status and parameter values ​​of each stage are recorded. The collected process parameter data are organized according to time sequence and process stage to generate a complete dataset containing the parameter change trends of each stage. The complete dataset is preprocessed to remove outliers and missing values, and the processed complete dataset is stored in a structured format.

3. The method according to claim 2, characterized in that, In step S1, the parameter correlation strength index is obtained, including: The dataset is processed using an association rule mining algorithm to extract frequent itemsets between parameters of each process step; the support and confidence between process parameters are calculated based on the frequent itemsets to generate association rules between parameters; the degree of influence between parameters is quantified through the association rules to obtain the parameter association strength index.

4. The method according to claim 1, characterized in that, In step S2, the linkage parameter group is determined, including: A preset threshold is set as the judgment standard, and the magnitude of the parameter correlation strength index is compared with the preset threshold. If the parameter correlation strength index exceeds the preset threshold, it is determined that there is a linkage relationship between the corresponding process parameters. The process parameters with the linkage relationship are grouped into linkage parameter groups. The linkage parameter groups are classified and archived, and the correlation mode and strength value of the parameters in each group are recorded.

5. The method according to claim 4, characterized in that, In step S2, a defect association dataset is constructed, including: Extract defect occurrence data related to historical linkage parameter groups from historical production records; analyze the specific parameter value change trends of each historical linkage parameter group in the defect occurrence data; map the parameter value change trends to defect types to generate a defect association dataset containing defect features and parameter change relationships; clean and format the defect association dataset to obtain the final defect association dataset.

6. The method according to claim 1, characterized in that, In step S3, the defect location rules are determined, including: A defect prediction model is constructed by training the defect association dataset using the random forest algorithm. The parameter value characteristics of each historical linkage parameter group when the defect occurs are analyzed through the defect prediction model to extract potential defect causes. Based on the correspondence between the potential defect causes and the historical linkage parameter groups, defect location rules are generated. The defect location rules are verified and the model parameters are adjusted.

7. The method according to claim 6, characterized in that, In step S3, the positioning results are obtained, including: Collect real-time process parameter data of the current conductor production; obtain real-time linkage data group based on the real-time process parameter data, input the real-time linkage data group into the defect prediction model, obtain defect prediction results, when the defect prediction result is abnormal, match the defect type and corresponding linkage data group in the defect prediction result with the defect location rules; identify abnormal parameter values ​​through the matching analysis, determine whether the abnormal parameter values ​​meet the characteristics of known defect causes; generate defect location results based on the determination results.

8. The method according to claim 1, characterized in that, In step S4, the process adjustment path is determined, including: The defect location results are fed back to the dataset to update the process parameter records in the dataset; the abnormal process parameters in the associated parameter group corresponding to the defect location results are compared with the normal process parameter values ​​in the associated parameter group corresponding to the nearest analysis unit, and the adjustment direction of the abnormal process parameters is identified based on the comparison difference and the potential defect causes; the final process adjustment path is generated through the adjustment direction to guide the optimization of the production process.

9. A digital process flow modeling system for main transformation conductors, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: The data acquisition unit is used to acquire process parameter data in the entire wire production process, organize them according to time sequence and process links, and form a complete dataset after preprocessing; The correlation analysis unit is used to analyze the correlation patterns between process parameters based on the dataset, combined with conductor length segmentation and time delay correction, and to obtain parameter correlation strength index. The defect association unit is used to determine the linkage relationship between process parameters based on the comparison result of the parameter association strength index and the preset threshold, and to determine the linkage parameter group; extract defect occurrence data related to the historical linkage parameter group from historical production records, analyze the parameter value change trend, and construct a defect association dataset; The defect localization unit is used to construct a defect prediction model for the defect association dataset, determine defect localization rules, collect real-time process parameter data, extract real-time linkage data groups, and obtain localization results based on the real-time linkage data groups through the defect prediction model and the defect localization rules. The process optimization unit is used to update the process parameter records in the dataset according to the positioning results, compare the abnormal process parameters in the linkage parameter group corresponding to the positioning results with the normal process parameter values ​​in the most recent linkage parameter group, and determine the process adjustment path based on the comparison difference and the potential defect causes.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-8.

Images