Production data anomaly analysis method and system

Abstract

The invention provides a production data anomaly analysis method and system, and relates to the technical field of data processing. The method comprises the following steps: acquiring historical cable crimping process record data and dividing the data into a core sample data set and a boundary sample data set; determining a crimping reference state according to the core sample data set, performing state deviation feature extraction on the boundary sample data set, and generating a plurality of state deviation sequences of the boundary sample data set; constructing deviation conduction data sets corresponding to a plurality of crimping stages, training the deviation conduction data sets to obtain a plurality of deviation conduction analysis models, performing deviation conduction correction on the boundary sample data set to generate a conduction correction data set, and constructing a stage deviation evolution graph based on the crimping reference state and the conduction correction data set, and performing anomaly analysis on the acquired real-time cable crimping data through the stage deviation evolution graph to generate a production data anomaly analysis result. According to the invention, the stability and the intelligent level of crimping process quality monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for analyzing anomaly in production data. Background Art

[0002] Cable crimping is a key process in the production of power equipment, widely used in the internal connection manufacturing of high- and low-voltage switchgear, box-type substations, and other equipment. The quality of crimping directly impacts the safety and stability of electrical connections. Some technologies, based on static threshold judgment and standard template matching, analyze key parameters in process data item by item. These methods essentially statically slice the crimping process, dividing the continuously changing data into discrete points or segments for judgment.

[0003] Cable crimping, however, is a process involving gradual pressure buildup and continuous evolution of contact conditions, resulting in highly process-dependent parameter changes. Deviations at a single point or within a localized timeframe may not fully reflect the overall evolution of the crimping state, leading to missed detection of potential anomalies. In this continuous process, early deviations can be partially offset or compensated for by subsequent steps due to factors such as equipment response and material deformation, resulting in the overall curve appearing normal despite a systematic shift. This type of in-process compensation is difficult to detect using traditional point-by-point analysis methods. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a production data anomaly analysis method and system, which can identify the real offset characteristics in the continuous processing of cable crimping, realize early identification of abnormal trends, and reveal potential risks from the overall evolution trend.

[0005] A first aspect of the present invention provides a method for analyzing production data anomalies, comprising: Obtain historical cable crimping process record data on the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set; Determine the crimping reference state based on the core sample data set, extract the state deviation feature of the boundary sample data set based on the crimping reference state, and generate the state offset sequence corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages; Constructing deviation conduction data sets corresponding to multiple crimping stages, training multiple deviation conduction analysis models based on the deviation conduction data sets, and performing deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set; Based on the crimping reference state and conduction correction data set, a stage deviation evolution map is constructed. The collected real-time cable crimping data is analyzed for anomalies through the stage deviation evolution map to generate production data anomaly analysis results.

[0006] Preferably, performing state deviation feature extraction on the boundary sample data set according to the crimping reference state includes: The crimping reference state includes reference crimping state sequences corresponding to multiple crimping stages in the cable crimping process for producing target power equipment, including constructing core crimping state sequences corresponding to each set of core sample data in the multiple crimping stages in the core sample data set, and fusing the multiple core crimping state sequences in the crimping stage to generate a reference crimping state sequence in the crimping stage; A boundary crimping state sequence corresponding to each group of boundary sample data in multiple crimping stages in the boundary sample data set is constructed, and deviation calculation is performed on the boundary crimping state sequence based on the reference crimping state sequence to generate multiple state offset sequences of the boundary sample data.

[0007] Preferably, a plurality of deviation conduction data sets corresponding to the respective crimping stages are constructed, and a plurality of deviation conduction analysis models are obtained by training based on the deviation conduction data sets, including: Calculating multiple stage state deviation values ​​for each group of boundary sample data based on the state offset sequence, determining the deviation conduction path of each of the multiple crimping stages except the first crimping stage, extracting multiple groups of conduction sample data for each deviation conduction path based on the multiple stage state deviation values, and generating multiple deviation conduction data sets; A deviation conduction analysis model is constructed for each deviation conduction data set according to the deviation conduction path, wherein the deviation conduction analysis model is a linear regression model. The deviation conduction analysis model is trained using the deviation conduction data set to obtain deviation conduction analysis models corresponding to the remaining crimping stages except the first crimping stage in multiple crimping stages.

[0008] Preferably, constructing a phase deviation evolution map based on the crimping reference state and the conduction correction data set includes: Correcting multiple sets of boundary sample data according to multiple deviation conduction analysis models to generate conduction correction data sets including multiple sets of conduction correction sample data, determining a corrected crimping state sequence for each set of conduction correction sample data, and determining a stage direction feature of each corrected crimping state sequence at each crimping stage based on a crimping reference state; Determine the stage deviation range of each crimping stage according to the multiple stage direction characteristics and the modified crimping state sequence, determine multiple state deviation intervals of the crimping stage according to the stage deviation range, and use the multiple state deviation intervals as state deviation nodes of each crimping stage in the stage deviation evolution map; Multiple stage deviation evolution paths are extracted from the conduction correction dataset. Based on the multiple stage deviation evolution paths, multiple deviation evolution edges about the state deviation nodes are determined, and the evolution weight of each deviation evolution edge is calculated to construct a stage deviation evolution graph about the multiple state deviation nodes.

[0009] Preferably, the collected real-time cable crimping data is analyzed for anomalies through the stage deviation evolution map to generate production data anomaly analysis results, including: Multiple real-time deviation evolution paths are extracted from the real-time cable crimping data, and the global deviation evolution weight of each real-time deviation evolution path is calculated according to the stage deviation evolution map. Multiple real-time deviation evolution paths are marked as abnormal based on the preset deviation evolution threshold, and the path proportion parameters marked as abnormal evolution paths in the real-time cable crimping data are determined to generate production data anomaly analysis results for the real-time cable crimping data.

[0010] Preferably, multiple groups of boundary sample data are corrected according to multiple deviation conduction analysis models to generate a conduction correction data set including multiple groups of conduction correction sample data, including: The multiple stage state deviation values ​​of each set of boundary sample data are input into the corresponding deviation conduction analysis model to generate the corrected state deviation values ​​corresponding to the stage state deviation values, and the conduction correction sample data corresponding to each set of boundary sample data are obtained to generate a conduction correction data set including multiple sets of conduction correction sample data.

[0011] A second aspect of the present invention provides a production data anomaly analysis system, which is used to implement the above-mentioned production data anomaly analysis method, comprising: A crimping data acquisition module is used to acquire historical cable crimping process record data about the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set; The state offset analysis module is used to determine the crimping reference state based on the core sample data set, extract the state deviation features of the boundary sample data set based on the crimping reference state, and generate the state offset sequence corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages; The deviation conduction analysis module is used to construct deviation conduction data sets corresponding to multiple crimping stages, train multiple deviation conduction analysis models based on the deviation conduction data sets, and perform deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set; The production data anomaly analysis module is used to construct a stage deviation evolution map based on the crimping reference state and conduction correction data set, perform anomaly analysis on the collected real-time cable crimping data through the stage deviation evolution map, and generate production data anomaly analysis results.

[0012] The present invention has the following beneficial effects: Aiming at the temporal continuity and multi-stage evolution characteristics of the cable crimping process in the production process of power equipment, the present invention proposes an anomaly detection mechanism that integrates deviation conduction modeling and deviation evolution spectrum analysis. By constructing a conduction model of inter-stage errors, the interference of error compensation effects between different crimping stages on anomaly identification is eliminated, and more representative real deviation features are extracted. The state offset evolution spectrum is further constructed in combination with historical representative sample data, which realizes the precise mapping of multi-stage deviation paths and anomaly weight analysis. It makes a great breakthrough in the traditional anomaly judgment method based on fixed thresholds or single-stage indicators, supports trend identification and precise early warning of potential abnormal states in real-time crimping data, and improves the stability and intelligence level of crimping process quality monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a method for analyzing production data anomalies according to one embodiment of the present invention is provided.

[0014] Figure 2 A schematic diagram of the structure of a production data anomaly analysis system provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] One embodiment of the present invention provides a method for analyzing production data anomalies, see Figure 1 , the method comprises the following steps: Step S1: Obtain historical cable crimping process record data on the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set.

[0017] Specifically, the historical cable crimping process record data included in the historical crimping process record data involves time series data of various process parameters collected online by sensor equipment during the cable crimping process, such as real-time change data of the pressure applied during the crimping process, real-time displacement data of the mold or crimping process, instantaneous movement speed data of the crimping head, etc. During the crimping process of some products, data such as mold temperature changes in the crimping equipment can also be collected to avoid interference caused by thermal expansion and contraction. The historical crimping quality test data included can be obtained by automatic or manual detection and recording after the crimping is completed, reflecting the performance quality of the finished crimped product, such as the pull-out force that characterizes the connection strength between the terminal and the cable, the contact resistance that characterizes the conductive performance, and the crimping height that characterizes whether the shaping meets the standards. The collected data is standardized and aligned, such as after normalization, for subsequent modeling and analysis.

[0018] For the pre-processed historical cable crimping process record data, it is divided according to the crimping quality to construct representative samples with excellent crimping effects and small deviations. Technical personnel or pre-set quality thresholds can be used to determine multiple groups of representative core sample data from a large amount of sample data to obtain a core sample data set for constructing a reference state. The remaining data is marked as boundary sample data to obtain a boundary sample data set, which contains normal samples that have a certain offset but are still within an acceptable range, and is used to capture edge changes and deviation conduction laws. That is, the collected historical cable crimping process record data can specifically be the data corresponding to products that meet the quality requirements during the production process. It is divided based on the crimping quality to obtain samples that are closer to the process boundary area, which helps to model the dynamic evolution process of the crimping state gradually deviating from the target requirements.

[0019] Step S2: Determine the crimping reference state based on the core sample data set, extract state deviation features of the boundary sample data set based on the crimping reference state, and generate state offset sequences corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages.

[0020] Specifically, the crimping reference state is constructed by extracting the average state of multiple crimping stages as the central benchmark based on multiple representative samples in the core sample data set, including the reference crimping state sequences corresponding to multiple crimping stages in the cable crimping process for producing target power equipment.

[0021] Among them, the crimping stage can be a plurality of typical stages obtained by dividing according to process characteristics, such as initial contact, plastic compression, crimping shaping and other stages, or it can be divided into multiple stages according to the crimping time, etc. Those skilled in the art can make reasonable divisions according to the actual production process. After determining the multiple crimping stages of the cable crimping process for producing the target power equipment, the core crimping state sequence corresponding to each group of core sample data in different crimping stages is extracted. First, the core crimping state sequence corresponding to each group of core sample data in the core sample data set in multiple crimping stages is constructed, which is expressed as a time series vector group composed of the values ​​of multiple sampling points in different parameter dimensions in the crimping stage, and then the multiple core crimping state sequences of each crimping stage are fused, for example, by calculating the mean, to generate a reference crimping state sequence corresponding to each crimping stage to construct a crimping reference state.

[0022] Furthermore, a comparative analysis of the crimping process is performed on the boundary sample data set using multiple reference crimping state sequences as a benchmark. By comparing the differences between the process states of each set of boundary sample data and the reference states, the state deviations at multiple crimping stages of the crimping process are extracted. For example, boundary crimping state sequences corresponding to the boundary sample data at each crimping stage are constructed. Deviation calculations, such as taking differences, are performed on the boundary crimping state sequences based on the reference crimping state sequences, forming state offset sequences corresponding to multiple crimping stages for each set of boundary sample data, reflecting the degree and directionality of the offset relative to the reference state at each crimping stage.

[0023] Step S3: construct deviation conduction data sets corresponding to multiple crimping stages, train multiple deviation conduction analysis models based on the deviation conduction data sets, and perform deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set.

[0024] It is worth noting that in a typical continuous manufacturing process such as cable crimping, the crimping process is usually completed by the equipment under closed-loop control. Therefore, between different crimping stages, the deviations of different parameters such as displacement and pressure may affect each other. For example, after a deviation occurs in a certain stage, due to the feedback control system of the equipment, such as the crimping displacement reaches the setting but the pressure is too small, or the natural tendency of material deformation, the subsequent stages may automatically or passively make up for the deviation, resulting in the final overall error being insignificant. This kind of deviation transmission and compensation relationship is often ignored in the anomaly detection method based on single-point threshold judgment, and it is impossible to identify the state where the anomaly has not yet appeared but has begun to evolve. The present invention improves the analysis depth of potential anomalies by modeling the deviation transmission path between each stage.

[0025] Specifically, the state offset sequence of each stage in the boundary sample data set is used as input to construct a deviation conduction data set between stages, and a deviation conduction analysis model is trained for each stage to express that the deviation of this stage is affected by the deviation of the previous stage. Based on the trained model, the state offset sequence of the boundary sample data set is corrected to eliminate some compensable deviations that may be brought by the previous stage, retain the true residual offset of each stage, and generate a conduction correction data set to characterize the actual independent deviation state of each sample after eliminating the mutual influence.

[0026] As an optional implementation process, a deviation conduction dataset corresponding to multiple crimping stages is constructed. Based on the deviation conduction dataset, multiple deviation conduction analysis models are trained, including: Based on the state offset sequence, multiple stage state deviation values ​​of each group of boundary sample data are calculated, and the deviation conduction path of each crimping stage except the first crimping stage is determined. Based on the multiple stage state deviation values, multiple groups of conduction sample data of each deviation conduction path are extracted to generate multiple deviation conduction data sets.

[0027] In this embodiment, for each state offset sequence, which involves deviation values ​​of multiple parameter dimensions at different sampling points, in order to simplify the feature dimension, the mean of each parameter of the state offset sequence, such as pressure and displacement, is calculated to obtain the reduced-dimensional state offset sequence, and the modulus of the reduced-dimensional sequence is calculated as the overall deviation intensity. While retaining the conduction relationship across stages, the actual deviation degree of the sample compared to the reference state in each crimping stage is quantified. Then, based on the concept of stage deviation conduction path, the modeling structure of deviation conduction is determined. Since the first crimping stage will not be affected by the previous stage, it usually starts from the second crimping stage. Each crimping stage will be affected by multiple upstream crimping stages. In this way, the deviation conduction path of the crimping stage is determined. For example, the fifth crimping stage will be affected by the deviations of the previous four crimping stages. After determining multiple deviation conduction paths, according to the crimping stages involved in the deviation conduction paths, multiple stage state deviation values ​​corresponding to these stages and belonging to the same group of boundary sample data are extracted to form one group of conduction sample data. Each group of conduction sample data includes the stage state deviation values ​​of multiple upstream crimping stages as input features and the stage state deviation value of the current crimping stage as the output target. In this way, the deviation conduction data set corresponding to each crimping stage is constructed.

[0028] A deviation conduction analysis model for each deviation conduction data set is constructed according to the deviation conduction path. The deviation conduction analysis model is trained using the deviation conduction data set to obtain deviation conduction analysis models corresponding to the remaining crimping stages except the first crimping stage in multiple crimping stages.

[0029] In this embodiment, the deviation transmission analysis model is preferably a linear regression model. Model parameter training is completed based on the deviation transmission dataset by minimizing the sum of squared residuals. This allows the deviation transmission analysis model to reflect the actual cross-stage deviation compensation, amplification, or transmission trends in historical samples. Ultimately, a deviation transmission analysis model corresponding to all crimping stages except the first crimping stage is obtained. This can be used to correct the state deviation sequence of boundary sample data, eliminating the masking effect caused by factors such as deviation transmission between stages, thereby obtaining a more essential and independent true deviation structure.

[0030] Step S4: construct a stage deviation evolution map based on the crimping reference state and the conduction correction data set, perform anomaly analysis on the collected real-time cable crimping data through the stage deviation evolution map, and generate a production data anomaly analysis result.

[0031] Specifically, for each sample in the conduction correction dataset, the actual deviation values ​​at multiple stages are represented symbolically. A stage-by-stage deviation evolution map is constructed, showing the deviation evolution relationship across multiple crimping stages. In practical applications, the collected cable crimping data undergoes the same deviation calculation and correction process, mapped to the map structure, and observed to see if its paths fall into high-risk areas, deviate from high-frequency main paths, or exhibit rare jumps. This generates anomaly analysis results for production data, enabling early warning of potential abnormal trends and ensuring quality assurance.

[0032] As an optional implementation, a phase deviation evolution map is constructed based on the crimp reference state and conduction correction datasets, including: Multiple groups of boundary sample data are corrected according to multiple deviation conduction analysis models to generate a conduction correction data set including multiple groups of conduction correction sample data, a corrected crimping state sequence of each group of conduction correction sample data is determined, and the stage direction characteristics of each corrected crimping state sequence in each crimping stage are determined based on the crimping reference state.

[0033] Among them, the process of correcting multiple groups of boundary sample data is specifically to input multiple stage state deviation values ​​of each group of boundary sample data into the deviation conduction analysis model of the corresponding crimping stage, so as to obtain a more essential corrected state deviation value of each group of boundary sample data in each crimping stage. For example, for the stage state deviation value of the third crimping stage, according to the corresponding deviation conduction analysis model, the stage state deviation values ​​of the first and second crimping stages are input into the trained linear regression model to generate the corrected state deviation value of the third crimping stage. In this way, multiple groups of boundary sample data are corrected to obtain multiple groups of conduction correction sample data, and each group of conduction correction sample data includes a corrected crimping state sequence composed of corrected state deviation values ​​based on multiple crimping stages.

[0034] Considering that the corrected state deviation value is a non-negative scalar value, if all deviations are non-negative, the fused deviation feature will only have "deviation magnitude" but no "deviation tendency." Therefore, a sign correction mechanism based on the center reference direction offset is introduced. Specifically, the dot product between the state deviation sequence of the boundary sample data and the corresponding reference crimping state sequence of the crimping stage is calculated, and the sign information of the dot product (positive or negative) is extracted as the stage direction feature of the crimping stage, which is used to describe whether the deviation represented by the corrected state deviation value is toward or away from the reference direction.

[0035] The stage deviation range of each crimping stage is determined according to multiple stage directional characteristics and the corrected crimping state sequence. Multiple state deviation intervals of the crimping stage are determined according to the stage deviation range. The multiple state deviation intervals are used as the state deviation nodes of each crimping stage in the stage deviation evolution map.

[0036] Among them, the corrected state deviation value in each corrected crimping state sequence is given a directional sign according to the stage directional characteristics, that is, a positive and negative sign is added, and the multiple corrected state deviation values ​​containing positive and negative signs in each crimping stage can be summarized to obtain the stage deviation range of the crimping stage. The stage deviation range of each crimping stage can be divided in a fixed step size or a fixed number, for example, into multiple state deviation intervals with equal step sizes, and the state deviation interval is used as the state deviation node in the stage deviation evolution map.

[0037] Multiple stage deviation evolution paths are extracted from the conduction correction dataset. Based on the multiple stage deviation evolution paths, multiple deviation evolution edges about the state deviation nodes are determined, and the evolution weight of each deviation evolution edge is calculated to construct a stage deviation evolution graph about the multiple state deviation nodes.

[0038] Specifically, for the stage deviation evolution path, the stage directional features of the boundary sample data can be integrated into the corrected crimping state sequence to determine the corrected crimping state sequence after the fusion directional features. The state deviation interval corresponding to each crimping stage, that is, the corresponding state deviation node in the stage deviation evolution map, can be used to determine the stage deviation evolution path of different boundary sample data. Then, based on multiple stage deviation evolution paths, the path transfer frequency of all samples can be counted, that is, the evolution weights between any two consecutive crimping stages and between any two adjacent state node pairs can be calculated. Finally, based on all state deviation nodes and the deviation evolution edges between them, a complete stage deviation evolution map is constructed as the joint evolution structure of multi-stage state deviations in the sample population.

[0039] As an optional implementation process, the collected real-time cable crimping data is analyzed for anomalies through the stage deviation evolution map to generate production data anomaly analysis results, including: Multiple real-time deviation evolution paths are extracted from the real-time cable crimping data. The global deviation evolution weight of each real-time deviation evolution path is calculated according to the stage deviation evolution map. Multiple real-time deviation evolution paths are marked as abnormal based on the preset deviation evolution threshold.

[0040] Specifically, similar to extracting the stage deviation evolution paths of multiple groups of boundary sample data in the boundary sample data set, the real-time deviation evolution paths of the real-time cable crimping data corresponding to the multiple groups of real-time sample data are constructed, each real-time deviation evolution path is mapped to the stage deviation evolution map, and the multiple evolution weights involved in the real-time deviation evolution path are determined. The average of the multiple evolution weights is calculated as the global deviation evolution weight of the real-time deviation evolution path, so as to comprehensively reflect the rarity or deviation of the path as a whole in the historical evolution map.

[0041] Then, according to the preset deviation evolution threshold, each real-time deviation evolution path is marked as abnormal. For example, if the global deviation evolution weight of a certain path is lower than the preset deviation evolution threshold, it will be judged as an abnormal evolution path. The preset deviation evolution threshold can be reasonably set according to the data in the actual production process. For example, if 100 consecutive products meet the standards, multiple consecutive products can be selected as representatives, and the global deviation evolution weight of each product can be calculated. The average of the global deviation evolution weights of these products is used as the preset deviation evolution threshold. That is, if no abnormality occurs for a long time in the continuous generation process, it means that the overall production status during this period is relatively stable and can be used as a representative of normal production. Those skilled in the art can reasonably set the preset deviation evolution threshold according to actual needs.

[0042] The system further identifies the abnormal evolution paths marked in the real-time cable crimping data, calculates the proportion of these abnormal paths, and generates path proportion parameters to reflect the overall volatility and abnormal trend of the current crimping process state. Finally, based on the single-path abnormality analysis and the overall path abnormality proportion parameters, it outputs the production data abnormality analysis results for the current real-time cable crimping data. In other words, if there are many abnormal state evolution processes in consecutive products, even if there are no obvious abnormalities overall, there is a high probability that potential abnormal factors will cause the product to gradually deviate from the normal production state. For this potential risk, it may be necessary to issue an early warning, etc. Ultimately, this method provides real-time monitoring and auxiliary decision-making basis for the quality control of the crimping process.

[0043] Based on the same concept of the above-mentioned production data anomaly analysis method, one embodiment of the present invention also provides a production data anomaly analysis system, see Figure 2 , the system comprises: A crimping data acquisition module is used to acquire historical cable crimping process record data about the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set; The state offset analysis module is used to determine the crimping reference state based on the core sample data set, extract the state deviation features of the boundary sample data set based on the crimping reference state, and generate the state offset sequence corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages; The deviation conduction analysis module is used to construct deviation conduction data sets corresponding to multiple crimping stages, train multiple deviation conduction analysis models based on the deviation conduction data sets, and perform deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set; The production data anomaly analysis module is used to construct a stage deviation evolution map based on the crimping reference state and conduction correction data set, perform anomaly analysis on the collected real-time cable crimping data through the stage deviation evolution map, and generate production data anomaly analysis results.

[0044] The foregoing description is merely a detailed description of the present invention, which is intended to enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Portions not described in detail in this specification are well known to those skilled in the art.

Claims

1. A method for analyzing abnormality in production data, characterized in that: include: Obtain historical cable crimping process record data on the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set; Determine the crimping reference state based on the core sample data set, extract the state deviation feature of the boundary sample data set based on the crimping reference state, and generate the state offset sequence corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages; Constructing deviation conduction data sets corresponding to multiple crimping stages, training multiple deviation conduction analysis models based on the deviation conduction data sets, and performing deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set; Based on the crimping reference state and conduction correction data set, a stage deviation evolution map is constructed. The collected real-time cable crimping data is analyzed for anomalies through the stage deviation evolution map to generate production data anomaly analysis results.

2. A method for analyzing production data anomalies according to claim 1, characterized in that: The state deviation feature extraction of the boundary sample data set is performed based on the crimping reference state, including: The crimping reference state includes reference crimping state sequences corresponding to multiple crimping stages in the cable crimping process for producing target power equipment, including constructing core crimping state sequences corresponding to each set of core sample data in the multiple crimping stages in the core sample data set, and fusing the multiple core crimping state sequences in the crimping stage to generate a reference crimping state sequence in the crimping stage; A boundary crimping state sequence corresponding to each group of boundary sample data in multiple crimping stages in the boundary sample data set is constructed, and deviation calculation is performed on the boundary crimping state sequence based on the reference crimping state sequence to generate multiple state offset sequences of the boundary sample data.

3. A method for analyzing production data anomalies according to claim 2, characterized in that: Construct deviation conduction datasets corresponding to multiple crimping stages. Based on the deviation conduction datasets, multiple deviation conduction analysis models are trained, including: Calculating multiple stage state deviation values ​​for each group of boundary sample data based on the state offset sequence, determining the deviation conduction path of each of the multiple crimping stages except the first crimping stage, extracting multiple groups of conduction sample data for each deviation conduction path based on the multiple stage state deviation values, and generating multiple deviation conduction data sets; A deviation conduction analysis model is constructed for each deviation conduction data set according to the deviation conduction path, wherein the deviation conduction analysis model is a linear regression model. The deviation conduction analysis model is trained using the deviation conduction data set to obtain deviation conduction analysis models corresponding to the remaining crimping stages except the first crimping stage in multiple crimping stages.

4. A method for analyzing production data anomalies according to claim 3, characterized in that: The phase deviation evolution map is constructed based on the crimp reference state and conduction correction dataset, including: Correcting multiple sets of boundary sample data according to multiple deviation conduction analysis models to generate conduction correction data sets including multiple sets of conduction correction sample data, determining a corrected crimping state sequence for each set of conduction correction sample data, and determining a stage direction feature of each corrected crimping state sequence at each crimping stage based on a crimping reference state; Determine the stage deviation range of each crimping stage according to the multiple stage direction characteristics and the modified crimping state sequence, determine multiple state deviation intervals of the crimping stage according to the stage deviation range, and use the multiple state deviation intervals as state deviation nodes of each crimping stage in the stage deviation evolution map; Multiple stage deviation evolution paths are extracted from the conduction correction dataset. Based on the multiple stage deviation evolution paths, multiple deviation evolution edges about the state deviation nodes are determined, and the evolution weight of each deviation evolution edge is calculated to construct a stage deviation evolution graph about the multiple state deviation nodes.

5. A method for analyzing production data anomalies according to claim 4, characterized in that: The collected real-time cable crimping data is analyzed for anomalies using the stage deviation evolution graph to generate production data anomaly analysis results, including: Multiple real-time deviation evolution paths are extracted from the real-time cable crimping data, and the global deviation evolution weight of each real-time deviation evolution path is calculated according to the stage deviation evolution map. Multiple real-time deviation evolution paths are marked as abnormal based on the preset deviation evolution threshold, and the path proportion parameters marked as abnormal evolution paths in the real-time cable crimping data are determined to generate production data anomaly analysis results for the real-time cable crimping data.

6. A method for analyzing production data anomalies according to claim 4, characterized in that: The plurality of groups of boundary sample data are corrected according to the plurality of deviation conduction analysis models to generate a conduction correction data set including the plurality of groups of conduction correction sample data, including: The multiple stage state deviation values ​​of each set of boundary sample data are input into the corresponding deviation conduction analysis model to generate the corrected state deviation values ​​corresponding to the stage state deviation values, and the conduction correction sample data corresponding to each set of boundary sample data are obtained to generate a conduction correction data set including multiple sets of conduction correction sample data.

7. A production data anomaly analysis system, characterized in that: The system is used to implement the production data anomaly analysis method according to any one of claims 1 to 6, comprising: A crimping data acquisition module is used to acquire historical cable crimping process record data about the production of target power equipment, including historical crimping process record data and historical crimping quality test data, and divide the historical cable crimping process record data into a core sample data set and a boundary sample data set; The state offset analysis module is used to determine the crimping reference state based on the core sample data set, extract the state deviation features of the boundary sample data set based on the crimping reference state, and generate the state offset sequence corresponding to each group of boundary sample data in the boundary sample data set in multiple crimping stages; The deviation conduction analysis module is used to construct deviation conduction data sets corresponding to multiple crimping stages, train multiple deviation conduction analysis models based on the deviation conduction data sets, and perform deviation conduction correction on the boundary sample data sets using the trained deviation conduction analysis models to generate a conduction correction data set; The production data anomaly analysis module is used to construct a stage deviation evolution map based on the crimping reference state and conduction correction data set, perform anomaly analysis on the collected real-time cable crimping data through the stage deviation evolution map, and generate production data anomaly analysis results.

Images