Terminal connector multi-station data acquisition and analysis method and platform

Abstract

The invention discloses a terminal connector multi-station data acquisition and analysis method and platform, and relates to the technical field of data acquisition and analys.The method comprises the steps that a unified machining time axis of a terminal connector is established, and a multi-station monitoring data set is established; state abnormity identification based on stations is carried out, and station abnormity is established; configuring a dynamic state chain by using a multi-station monitoring data set and a processing time axis; performing residual analysis according to the dynamic state chain and the healthy connector track, and establishing an abnormal focus point; configuring a behavior causal database between stations, and establishing a backtracking positioning result; and generating a terminal connector quality detection result. According to the invention, the technical problems of non-uniform multi-station data acquisition, low abnormal positioning precision and low quality detection efficiency in the production process of the terminal connector in the prior art are solved, and the technical effects of realizing multi-station data acquisition in the whole production process of the terminal connector and improving the abnormal positioning precision and the quality detection efficiency in the production process are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data acquisition and analysis, and in particular to a terminal connector multi-station data acquisition and analysis method and platform. Background Art

[0002] In the multi-station production process of terminal connectors, due to the lack of a unified time base for data collection at each station, it is difficult to achieve full-process production tracking based on a unique ID, resulting in the inability to effectively associate multi-station monitoring data and difficulty in establishing a complete monitoring data set; at the same time, the existing technology lacks specificity in identifying abnormalities in station status, and is unable to accurately locate abnormal stations, and has not built a dynamic status chain to reflect the status changes of the production process, making the abnormality analysis lack of systematicity; in addition, due to the lack of a behavioral causal database between stations, it is difficult to conduct retrospective analysis of real abnormalities based on the abnormal focus point, resulting in inaccurate quality inspection results, which cannot meet the efficiency and accuracy requirements for abnormality positioning and quality inspection in the terminal connector production process.

[0003] The existing technology solves the technical problems of inconsistent multi-station data collection in the terminal connector production process, low abnormal positioning accuracy and low quality inspection efficiency. Summary of the Invention

[0004] The present application provides a terminal connector multi-station data collection and analysis method and platform, which is used to solve the technical problems in the prior art of non-uniform multi-station data collection in the terminal connector production process, low abnormal positioning accuracy and low quality inspection efficiency.

[0005] In view of the above problems, the present application provides a terminal connector multi-station data collection and analysis method and platform.

[0006] A first aspect of the present application provides a method for collecting and analyzing multi-station data of a terminal connector, the method comprising: Establish a unified processing timeline for terminal connectors, conduct terminal connector production tracking based on a unique binding ID based on the unified processing timeline, and establish a multi-station monitoring data set; perform station-based state anomaly identification on the multi-station monitoring data set to establish station anomalies; configure a dynamic state chain using the multi-station monitoring data set and the processing timeline; perform residual analysis based on the dynamic state chain and the healthy connector trajectory to establish anomaly focus points; configure a behavior causal database between stations based on the production process, conduct real anomaly backtracking analysis based on the behavior causal database and the anomaly focus points, and establish a backtracking positioning result; generate terminal connector quality inspection results based on the backtracking positioning results and the station anomaly.

[0007] A second aspect of the present application provides a terminal connector multi-station data acquisition and analysis platform, the platform comprising: A monitoring data set establishment module is used to establish a unified processing timeline for terminal connectors, and based on the unified processing timeline, terminal connector production tracking based on a unique binding ID is performed to establish a multi-station monitoring data set; a station anomaly establishment module is used to perform station-based state anomaly identification on the multi-station monitoring data set and establish station anomalies; a dynamic state chain configuration module is used to configure a dynamic state chain using the multi-station monitoring data set and the processing timeline; an anomaly focus point establishment module is used to perform residual analysis based on the dynamic state chain and the healthy connector trajectory to establish an anomaly focus point; a backtracking positioning result establishment module is used to configure a behavior causal database between stations based on the production process, perform real anomaly backtracking analysis based on the behavior causal database and the anomaly focus point, and establish a backtracking positioning result; a quality inspection result generation module is used to generate terminal connector quality inspection results based on the backtracking positioning results and the station anomaly.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: A unified processing timeline for terminal connectors was established, along with a multi-station monitoring data set. The multi-station monitoring data set was used to identify state anomalies based on the workstations, establishing workstation anomalies. A dynamic state chain was configured using the multi-station monitoring data set and the processing timeline. Residual analysis was performed to establish anomaly focal points. A behavioral causal database between workstations was configured based on the production process, and actual anomaly backtracking analysis was performed to establish backtracking location results. Terminal connector quality inspection results were generated based on the backtracking location results and the workstation anomalies. This approach achieved the technical effect of enabling multi-station data collection for the entire terminal connector production process, improving the accuracy of anomaly location in the production process and the efficiency of quality inspection. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0010] Figure 1 A schematic flow chart of a method for collecting and analyzing multi-station data of a terminal connector provided in an embodiment of the present application; Figure 2 This is a structural diagram of a terminal connector multi-station data acquisition and analysis platform provided in an embodiment of the present application.

[0011] Explanation of the accompanying symbols: monitoring data set establishment module 10, work station anomaly establishment module 20, dynamic state chain configuration module 30, anomaly focus point establishment module 40, backtracking positioning result establishment module 50, quality detection result generation module 60. DETAILED DESCRIPTION

[0012] The present application provides a terminal connector multi-station data collection and analysis method and platform, which is used to solve the technical problems in the prior art of non-uniform multi-station data collection in the terminal connector production process, low abnormal positioning accuracy and low quality inspection efficiency.

[0013] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, the present application provides a terminal connector multi-station data collection and analysis method, the method comprising: Step S100: establishing a unified processing timeline for terminal connectors, performing terminal connector production tracking based on a unique binding ID based on the unified processing timeline, and establishing a multi-station monitoring data set.

[0015] Specifically, establishing a unified processing timeline for terminal connectors requires calibrating the time scales of stamping, crimping, and injection molding processes to millisecond-level accuracy, based on the process timing of each workstation. This forms a time benchmark that runs through the entire processing process. Based on this timeline, each terminal connector is assigned a unique binding ID. Through carriers such as RFID tags or visual identification codes, real-time data such as equipment operating parameters, material batches, and process indicators are collected at each processing link. These data are then associated with the corresponding IDs in timeline order, thereby integrating processing information from multiple workstations such as stamping, crimping, and injection molding. This establishes a multi-workstation monitoring data set covering production data throughout the entire lifecycle, enabling accurate production tracking from raw materials to finished products.

[0016] Step S200: performing workstation-based status anomaly identification on the multi-workstation monitoring data set to establish workstation anomalies.

[0017] Specifically, after splitting the multi-station monitoring data set according to the station dimension, unsupervised anomaly recognition models (such as isolation forests) are constructed for the process characteristics of different stations such as stamping, crimping, and injection molding. The monitoring data of each station (including stamping pressure curve, crimping height fluctuation value, injection molding temperature parameters, etc.) are input into the corresponding model. By calculating the degree of deviation of the data points from the normal distribution, the abnormal state that exceeds the preset threshold is identified, and then the station abnormality record containing information such as the time of abnormality occurrence, parameter deviation range and abnormality type is established, providing accurate station-level abnormality data support for subsequent abnormality focusing and retrospective analysis.

[0018] Step S300: configuring a dynamic state chain using the multi-station monitoring data set and the processing timeline.

[0019] Specifically, a dynamic state chain is configured based on the terminal connector production process sequence (e.g., stamping, crimping, and injection molding) using processing data such as equipment operating parameters and material batches from each workstation in a multi-workstation monitoring dataset, along with a unified processing timeline. The data from each workstation is linked together along the timeline, with each workstation serving as a state chain node. The input materials, control parameters, quality indicators, and other attributes of the corresponding process, along with a timestamp, are recorded. Through process linkage, a chain structure is formed that reflects the evolution of the product's processing state. This provides a time-series state evolution reference for subsequent residual analysis and lays a dynamic time-series foundation for establishing anomaly focus points and conducting backtracking analysis of true anomalies.

[0020] Step S400: Perform residual analysis based on the dynamic state chain and healthy connector trajectory to establish an abnormal focus point.

[0021] Specifically, the dynamic state chain and the healthy connector trajectory, established by clustering historically normal workpieces and updating them through a time window sliding update (combined with confidence weight compensation for production parameter adjustment nodes), are jointly input into the twin residual channel. The representation extractor within the channel extracts and fuses features from source data such as time series, images, and numerical values, outputting low-dimensional embedding vector pairs. The residual calculation layer then calculates the difference between these vector pairs, and the discriminator outputs an anomaly score. After normalization of the anomaly score, workstation mapping is performed. After threshold screening and focused analysis of adjacent workstations and time-continuous anomalies, a focused segment is constructed and anomaly focus points are generated, enabling precise location of the critical workstations and time nodes where the anomaly occurred.

[0022] Step S500: Based on the behavior causal database between the production process configuration workstations, a real anomaly backtracking analysis is performed according to the behavior causal database and the anomaly focus point to establish a backtracking positioning result.

[0023] Specifically, a process flow analysis is conducted based on the production process, extracting attributes such as input / output materials, control parameters (such as crimping pressure), and quality indicators (such as insertion and extraction force) for each workstation. Through workstation dependency analysis, primary causal data such as material flow and parameter transfer are constructed. Simultaneously, an expert rule library is invoked to analyze the mapping rules between process deviations (such as injection temperature exceeding limits) and quality results (such as housing cracking), forming secondary causal data. The two are integrated into a behavioral causal database. Based on the anomaly focus, the database is queried for a set of predecessor paths. Through backtracking verification, the confidence level of each path is calculated. High-confidence paths are selected to complete a backtracking analysis of the actual anomaly, identifying the root cause of the anomaly (such as mold wear) and establishing a backtracking location result that includes the specific faulty workstation and parameter deviation type.

[0024] Step S600: generating a terminal connector quality inspection result based on the backtracking positioning result and the workstation abnormality.

[0025] Specifically, the terminal connector quality inspection results are generated based on the root cause of the anomaly (such as crimping die wear) and the corresponding faulty workstation and parameter deviation type in the retrospective positioning results, combined with information such as the time of anomaly occurrence and parameter deviation range in the anomaly records of each workstation. First, the specific type of quality anomaly (such as poor terminal crimping), the workstation where it occurred, and the time window range (the window size is positively correlated with the quality anomaly value) are recorded, and a list of workstation features to focus on in the time window is generated. At the same time, using the retrospective positioning results and workstation anomaly data, an anomaly time evolution prediction model is used to analyze the anomaly development trend, establish auxiliary certification anomalies, and compensate and correct the quality inspection results. Finally, a quality inspection result is formed that includes the anomaly type, the workstation where it occurred, the time range, and the corrected inspection conclusion. This is used to guide workstation monitoring management and quality improvement in terminal connector production.

[0026] In one possible implementation method, step S400 further includes: Step S410: Input the dynamic state chain and healthy connector trajectory into the twin residual channel.

[0027] Step S420: Use the representation extractor in the twin residual channel to extract features of time series, image, and numerical source data respectively, perform feature fusion, and output low-dimensional embedding vector pairs.

[0028] Step S430: Use the residual calculation layer of the twin residual channel to perform residual calculation of the low-dimensional embedding vector pair, and output the anomaly score through the discriminator.

[0029] Step S440: establishing an abnormal focus point according to the abnormality score.

[0030] Specifically, the dynamic state chain (i.e., the state evolution chain formed by connecting the processing data and timestamps of each workstation in series according to the process sequence of stamping, crimping, injection molding, etc.) and the healthy connector trajectory (a standard trajectory constructed by clustering historical normal workpieces, sliding updates of time windows, and confidence weight compensation of nodes based on production parameter adjustment) are jointly input into the twin residual channel to provide a benchmark input for subsequent feature comparison and analysis of multi-source data.

[0031] Utilizing the representation extractor within the twin residual channel, corresponding algorithms are used to extract features from time series data (such as time-varying curves of crimping pressure and injection temperature), image data (such as appearance images obtained from visual inspection of terminals), and numerical data (such as the specific values ​​of processing parameters at each workstation) in the dynamic state chain and healthy connector trajectory. An LSTM network is used to capture the temporal dependencies of the time series, a CNN network is used to extract texture and shape features of the image, and statistical functions are used to calculate distributional features such as the mean and variance of the numerical data. After extraction, a weighted fusion layer is used to reduce the dimensionality and remove redundancy from the multi-source features. The final output is a low-dimensional embedding vector pair representing the dynamic state chain and healthy trajectory, providing standardized feature input for residual calculation.

[0032] The residual computation layer of the twin residual channel performs element-by-element interpolation on the low-dimensional embedding vector pairs of the dynamic state chain and the healthy connector trajectory, generating a residual vector representing the characteristic deviation between them. The residual computation layer quantifies the differences between the vector pairs using cosine similarity, forming a numerical residual matrix. This residual matrix is ​​input to the discriminator (composed of a fully connected neural network). After multiple layers of nonlinear transformations, it outputs an anomaly score between 0 and 1. A higher score indicates a greater deviation from the healthy trajectory of the current state chain, providing a quantitative basis for establishing subsequent anomaly focus points.

[0033] When establishing anomaly focus points based on anomaly scores, the anomaly scores are first normalized and mapped to the interval [0, 1]. The normalized scores are then mapped to specific workstations and time nodes based on the corresponding relationship between the processing timeline and workstations. A threshold (e.g., 0.8) is set to filter out regions with high anomaly scores. This filtered anomaly score is then subjected to a focused analysis of adjacent workstations and temporally continuous anomalies. Specifically, the anomaly scores that appear consecutively in a time series at adjacent workstations are analyzed to identify anomaly segments with temporal correlation and workstation association, and then a focused segment is constructed. Finally, based on the key time points, corresponding workstations, and degree of anomaly in the focused segment, anomaly focus points are generated, enabling precise localization of anomalies in the terminal connector production process.

[0034] In one possible implementation method, step S440 further includes: Step S441: After normalizing the abnormality scores, the normalized abnormality scores are mapped to workstations.

[0035] Step S442: After threshold screening of the normalized anomaly scores, focus analysis of adjacent workstations and time-continuous anomalies is performed based on the workstation mapping to construct a focused segment.

[0036] Step S443: Generate abnormal focus points using the focus segments.

[0037] Specifically, the anomaly scores output by the twin residual channel are normalized and linearly mapped to the interval [0, 1] using a minimum-maximum normalization method to eliminate dimensional differences in data from different workstations. Then, based on the time-series correspondence between the unified processing timeline and each workstation, the normalized anomaly scores are mapped to specific workstations (such as stamping, crimping, and injection molding) and their corresponding processing time nodes. A three-dimensional correlation mapping is established between the anomaly scores, physical workstations, and time, providing a precise spatial positioning foundation for subsequent anomaly-focused analysis.

[0038] A reasonable threshold (e.g., 0.8) is set for the normalized anomaly score, and data with anomaly scores exceeding the threshold is filtered out, retaining time nodes with significant anomalies. Based on the workstation mapping results, the anomaly scores of adjacent workstations (e.g., crimping and injection molding) that appear consecutively in a time series are analyzed to identify anomaly segments with process logic relevance. For example, a pressure anomaly at the crimping station is followed by a temperature anomaly at the injection molding station, and the time interval is consistent with the production cycle. This allows for the construction of a focused segment that includes the anomaly station, time range, and score, highlighting the continuity of anomalies in the spatiotemporal dimensions and the causal relationship between the workstations.

[0039] When generating anomaly focus points from constructed focused segments, the team first extracts the set of abnormal workstations (such as crimping and injection molding) contained in the segment, along with the time interval of consecutive anomalies and the peak anomaly score. This information is then integrated into a structured anomaly focus point. This focus point clearly identifies the specific workstation where the anomaly occurred (using a unique binding ID to locate the specific device), the time window (the window size is positively correlated with the duration of the anomaly), and the severity of the anomaly (classified as mild, severe, etc. based on the peak score). It also associates the abnormal linkage relationships between adjacent workstations (for example, an abnormal crimping parameter causes subsequent quality fluctuations at the injection molding station). This transforms scattered anomaly segments into precise anomaly location coordinates, providing clear target nodes for subsequent retrospective analysis of real anomalies based on the behavioral causal database.

[0040] In one possible implementation method, step S500 further includes: Step S510: Perform process analysis based on the production process and extract workstation attributes, which include input and output materials, control parameters, and quality indicators.

[0041] Step S520: Performing a workstation dependency analysis based on the workstation attributes to establish first causal association data.

[0042] Step S530: calling an expert rule base according to the production process, performing rule item analysis of mapping process deviations and quality results according to the expert rule base, and establishing second causal association data.

[0043] Step S540: constructing the behavior causal database based on the first causal association data and the second causal association data.

[0044] Specifically, we conduct a process analysis based on the terminal connector production process (covering processes such as stamping, crimping, and injection molding), and extract station attributes for each station (such as stamping, crimping, and injection molding). Input materials refer to the raw materials or semi-finished products required for processing at each station, such as copper strips for stamping and semi-finished stamped terminals for crimping. Output materials are the products processed at that station, such as terminal blanks produced by stamping and finished terminals formed by injection molding. Control parameters are process adjustment parameters that ensure normal processing at the station, such as stamping pressure at stamping, crimping height at crimping, and injection temperature at injection molding. Quality indicators are standards for measuring the processing quality of the station, including terminal dimensional tolerance, insertion and extraction force, insulation resistance, etc. By combing through these four attributes, we form a structured set of station attributes, laying the foundation for subsequent station dependency analysis and the construction of causal association data.

[0045] Based on extracted workstation attributes (including input / output materials, control parameters, and quality indicators), a dependency analysis was conducted on each workstation (such as stamping, crimping, and injection molding) in the terminal connector production process. By analyzing the material flow chain (e.g., the output semi-finished product of the stamping workstation serves as the input material for the crimping workstation), parameter transfer logic (e.g., the pressure parameters of the crimping workstation affect the mold compatibility of the injection molding workstation), and quality indicator transmission paths (e.g., dimensional tolerance deviations at the stamping workstation can lead to abnormal insertion and extraction forces at the crimping workstation), a first-order causal relationship data reflecting the direct causal relationship between workstations was established. The influence intensity (e.g., material dependency, parameter transmission coefficient) and specific influence paths (e.g., "stamping die wear → terminal blank dimensional tolerance → increased crimping defect rate at the crimping workstation") from workstation A to workstation B were recorded in the form of a directed graph, forming a structured direct causal relationship network between workstations.

[0046] Based on production process requirements, a pre-built expert rule library is invoked. This rule library integrates the empirical insights of senior industry experts on the correlation between process deviations and quality outcomes in terminal connector production. These include rules such as "When the crimping temperature exceeds the standard range by ±5°C, the probability of terminal crimp cracking increases by 20%," and "Injection pressure fluctuations exceeding the rated value by 10% will increase the housing short-fill defect rate by 15%." These rule entries are systematically analyzed, and the mapping between process deviations (such as stamping die wear and abnormal injection temperature) and quality outcomes (such as terminal dimensional tolerances and reduced insulation performance) is structured. This clarifies the extent and path of impact of different process deviations on quality outcomes, thereby establishing secondary causal association data reflecting potential causal relationships between indirect workstations, providing expert-based logical reasoning support for the behavioral causal database.

[0047] Based on the acquired primary causal association data (covering direct dependencies between workstations) and secondary causal association data (including mapping rules between process deviations and quality results), a behavioral causal database is constructed. Using each workstation, input and output materials, control parameters, and quality indicators as foundational elements, and the direct associations derived from workstation dependency analysis and the mapping between process deviations and quality results in the expert rule base as the connecting logic, this data is structured and integrated to form a database that comprehensively reflects the causal relationships between each link in the terminal connector production process. This provides an accurate and complete causal relationship query basis for subsequent retrospective analysis of real anomalies based on anomaly focus points.

[0048] In one possible implementation method, step S500 further includes: Step S550: performing a predecessor path query based on the behavior causal database according to the abnormal focus point, and establishing a predecessor path set.

[0049] Step S560: Perform backtracking verification on the predecessor path set to establish backtracking verification trust.

[0050] Step S570: Use the backtracking verification trust to perform backtracking screening to complete the real abnormality backtracking analysis.

[0051] Specifically, starting with the anomaly focal point, a precursor path query is conducted in the behavioral causal database. Starting from the specific workstation or quality result where the anomaly occurred, all possible factors and paths that may have caused the anomaly are traced back according to the workstation dependencies recorded in the database and the mapping rules between process deviations and quality results. For example, if the anomaly focal point is determined to be a shell cracking problem at the injection molding workstation, then the behavioral causal database is queried for all precursor factors related to shell cracking, such as abnormal injection temperature, mold wear, and excessive raw material moisture. All these possible influencing paths are collected to form a precursor path set containing multiple potential causes and corresponding causal chains, providing comprehensive clues for subsequent verification and screening of the anomaly causes.

[0052] For each path in the precursor path set, retrospective verification is performed by cross-checking production execution data (such as pressure and temperature parameters collected in real time by equipment sensors), process specification documents (such as the standard range of parameters for each workstation), and historical quality records (such as defective product statistics within the same time period). For example, for the path of "stamping die wear → terminal size deviation → poor crimping," it is necessary to verify the stamping die replacement cycle records and wear inspection reports, and at the same time retrieve the real-time pressure curve of the crimping station and the terminal size sampling data to calculate the probability of association between each link in the path. Using a fuzzy logic algorithm, the verification results are quantified into a confidence value between 0 and 1. Paths with high data consistency and coherent process logic are assigned high confidence, while those with poor process logic are assigned low confidence. Ultimately, a verification system is formed that includes confidence scores for each path.

[0053] Using the established backtracking verification confidence level, the set of predecessor paths is screened. A confidence threshold (e.g., 0.6) is set, and paths with confidence levels above the threshold are retained, while paths with low confidence levels are eliminated. For example, if a path has a confidence level of 0.82 and another has a confidence level of 0.41, the former is retained and the latter is eliminated. In this way, the path most likely to cause an anomaly is screened from the numerous potential causes, and the root cause of the anomaly is identified. For example, if stamping die wear is determined to be the primary cause of the current terminal crimping failure, a true backtracking analysis of the anomaly is completed, providing an accurate basis for subsequent targeted improvement measures.

[0054] In one possible implementation method, step S200 further includes: Step S210: Establish an unsupervised anomaly recognition model for each workstation.

[0055] Step S220: input the multi-station monitoring data sets into the unsupervised anomaly recognition model respectively, and output the workstation anomaly.

[0056] Specifically, an unsupervised anomaly recognition model is established for each workstation using the Isolation Forest algorithm. This algorithm, based on the principle that anomalous data points are more easily isolated in feature space, constructs a binary tree of historical normal production data (e.g., parameters such as stamping pressure and crimping height) for each workstation (e.g., stamping and crimping). The degree of anomaly is assessed by calculating the average path length of each data point within the tree. Normal data points, due to their location in dense data areas, have shorter paths, while anomalous data points, due to their sparse distribution, have longer paths. When new real-time data from a workstation is input, the algorithm determines whether it is an anomaly based on a preset anomaly threshold (e.g., a path length exceeding 1.5 times the average level), thereby establishing a precise anomaly recognition model for each workstation.

[0057] The multi-station monitoring dataset is split according to station category and fed into the established unsupervised anomaly recognition model for the corresponding station. Taking the stamping station as an example, the real-time data collected from the station, such as stamping pressure and mold temperature, is fed into the isolation forest model for the stamping station. The model determines whether the data deviates from the normal distribution by calculating the path length of the data points and comparing it with a preset threshold. If the path length of the stamping pressure data exceeds the threshold within a certain time period, the result is output that the stamping station has an anomaly during that period. Similarly, the crimping height and crimping pressure data of the crimping station are fed into the crimping station model to identify abnormal conditions during the crimping process in real time, ultimately achieving automatic detection and output of anomalies in each station.

[0058] In one possible implementation method, step S400 further includes: Step S450: Clustering historical normal artifacts to establish representative trajectory groups.

[0059] Step S460: performing a time window sliding update of the representative trajectory group based on the trajectory evolution mechanism.

[0060] Step S470: extracting adjustment nodes of production parameters, establishing confidence weights according to the adjustment nodes, and using the confidence weights to compensate for time window sliding updates to establish the healthy connector trajectory.

[0061] Specifically, cluster analysis is performed on historically normal workpiece processing trajectory data (such as parameter time series and motion trajectories at each workstation). Using the K-means clustering algorithm, normal workpieces are divided into different groups based on trajectory similarity. Each group corresponds to a typical normal production trajectory pattern, thereby establishing representative trajectory groups. For example, the trajectories of parameters such as stamping pressure and displacement of historically normal workpieces at a stamping station are clustered to obtain typical normal trajectory groups under different production conditions, providing a standard reference model for subsequent anomaly detection.

[0062] Based on the trajectory evolution mechanism, a time window sliding update is implemented for the representative trajectory family. A fixed time window (e.g., in units of shifts or days) is set. When new normal workpiece machining trajectory data is collected, the time window is slid forward along the time axis, removing the earliest data within the window and incorporating the latest trajectory data to update the representative trajectory family. This method enables the representative trajectory family to adapt to dynamic changes in the production process, such as equipment wear and process parameter fine-tuning, ensuring that it always reflects the trajectory characteristics of the current normal production state, providing an accurate healthy trajectory reference for subsequent residual analysis.

[0063] Identify key adjustment nodes from the historical records of production parameters, such as equipment maintenance time and the moment when process parameters are reset. Assign confidence weights to each adjustment node based on the nature and impact of the adjustment. For example, parameter adjustments after major equipment maintenance are given higher weights, while smaller parameter adjustments are given lower weights. When performing a sliding time window update on the representative trajectory family, these confidence weights are used to compensate for the impact of new and old data. That is, a higher weight is given to new data after the adjustment node, and the weight of old data before the adjustment node is appropriately reduced. This constructs a healthy connector trajectory that accurately reflects the current production status, ensures that the trajectory is dynamically optimized as production conditions change, and provides a reliable normal trajectory benchmark for subsequent residual analysis.

[0064] In one possible implementation method, step S600 further includes: Step S610: Record the quality anomaly and generate a workstation feature focus of the quality anomaly. The workstation feature focus is a time window focus, and the window size of the time window is positively correlated with the quality anomaly value.

[0065] Step S620: Utilize the work station characteristics to monitor and manage the production of terminal connectors.

[0066] Specifically, when a quality anomaly is detected during the production of terminal connectors, detailed information about the anomaly is first recorded, including the specific time the anomaly occurred, the workstation involved, the type of anomaly (e.g., crimp height out-of-tolerance, unqualified dimensions, etc.), and the values ​​of related parameters. At the same time, a corresponding workstation feature attention is generated for the quality anomaly. This attention is reflected in the form of a time window, and the size of the time window is positively correlated with the quality anomaly value. If the quality anomaly value is large, indicating a high severity of the problem, the time window range is correspondingly expanded to cover a wider range of historical data, facilitating a comprehensive analysis of the production status before and after the anomaly. If the anomaly value is small, the time window is narrowed to focus on a more precise time period for monitoring. In this way, the quality anomaly is accurately located and targeted attention is focused on the time dimension.

[0067] The generated workstation feature attention (i.e., time window attention) is applied to workstation monitoring and management during the terminal connector production process. By focusing on production data within the time window (such as equipment operating parameters, material input and output status, quality inspection results, etc.), the operating status of the corresponding workstation within a specific time period is monitored in real time. For example, when the time window attention of a certain crimping workstation shows that the crimping pressure has fluctuated frequently and exceeded the normal range within the past two hours, an early warning is automatically triggered. Combined with the positive correlation between the time window size and outliers, the monitoring range is expanded to analyze the fluctuation trend, and then the staff is guided to adjust the crimping parameters or perform equipment maintenance, achieving precise control of the workstation production status and improving product quality stability and production efficiency.

[0068] In one possible implementation method, step S600 further includes: Step S630: Utilize the retrospective positioning result and the workstation anomaly to predict the anomaly time evolution and establish auxiliary authentication anomaly.

[0069] Step S640: Compensating the terminal connector quality inspection result based on the auxiliary authentication anomaly.

[0070] Specifically, using retrospective location results and workstation anomaly data, an LSTM (Long Short-Term Memory) network is employed. The root cause characteristics (such as equipment wear and parameter deviation) obtained from retrospective location and the historical time series data of workstation anomalies (such as anomaly indicator values ​​at each time point) are used as input to the LSTM model. Through the model's memory units and gating mechanism, the model learns the patterns and trends of anomaly characteristics over time, and then predicts the development of anomalies in future time periods (such as the magnitude of anomaly values ​​and the expansion of the impact range). This creates an auxiliary certification anomaly model containing anomaly evolution prediction information, providing a more accurate basis for subsequent quality inspection result compensation.

[0071] Compensation is applied to the terminal connector quality inspection results based on the predicted time evolution of anomalies contained in auxiliary certification anomalies (such as the anomaly's future development trend and potential impact range). For example, if an auxiliary certification anomaly indicates that a workstation anomaly will cause a 5% increase in product rejection rate within the next hour, this predicted potential rejection rate is incorporated into the current quality inspection results. The threshold for the inspection results is adjusted, or the number of rejected products detected is directly corrected. This ensures that the quality inspection results not only reflect current quality issues but also potential future quality risks, thereby providing a more comprehensive and accurate assessment of product quality status and providing a more forward-looking basis for production process control and quality improvement.

[0072] Embodiment 2 is based on the same inventive concept as the method for collecting and analyzing multi-station data of a terminal connector in the above embodiment. Figure 2As shown, the present application provides a terminal connector multi-station data acquisition and analysis platform. The platform and method embodiments in the present application are based on the same inventive concept. The platform includes: The monitoring data set establishing module 10 is used to establish a unified processing timeline for the terminal connector, perform terminal connector production tracking based on a unique binding ID based on the unified processing timeline, and establish a multi-station monitoring data set.

[0073] The workstation anomaly establishing module 20 is used to perform workstation-based state anomaly identification on the multi-workstation monitoring data set and establish workstation anomalies.

[0074] The dynamic state chain configuration module 30 is used to configure a dynamic state chain using the multi-station monitoring data set and the processing time axis.

[0075] The abnormal focus point establishing module 40 is used to perform residual analysis based on the dynamic state chain and the healthy connector trajectory to establish an abnormal focus point.

[0076] The backtracking location result establishing module 50 is used to configure the behavior causal database between workstations based on the production process, perform real anomaly backtracking analysis according to the behavior causal database and the anomaly focus point, and establish a backtracking location result.

[0077] The quality inspection result generating module 60 is configured to generate a terminal connector quality inspection result based on the backtracking positioning result and the workstation abnormality.

[0078] Furthermore, the platform is also used to implement the following functions: The dynamic state chain and healthy connector trajectory are input into the twin residual channel; the representation extractor in the twin residual channel is used to extract features of time series, image, and numerical source data respectively, and feature fusion is performed to output low-dimensional embedding vector pairs; the residual calculation layer of the twin residual channel is used to perform residual calculation of the low-dimensional embedding vector pairs, and the discriminator outputs the anomaly score; and an anomaly focus is established based on the anomaly score.

[0079] Furthermore, the platform is also used to implement the following functions: After normalizing the anomaly scores, the normalized anomaly scores are mapped to workstations; after threshold screening of the normalized anomaly scores, a focus analysis of adjacent workstations and time-continuous anomalies is performed based on the workstation mapping to construct a focus segment; and the focus segment is used to generate anomaly focus points.

[0080] Furthermore, the platform is also used to implement the following functions: Perform a process flow analysis based on the production process and extract workstation attributes, which include input and output materials, control parameters, and quality indicators; perform a workstation dependency analysis based on the workstation attributes and establish first causal association data; call an expert rule library based on the production process, perform rule entry analysis of mapping process deviations and quality results based on the expert rule library, and establish second causal association data; and construct the behavioral causal database based on the first causal association data and the second causal association data.

[0081] Furthermore, the platform is also used to implement the following functions: According to the abnormal focus point, a predecessor path query is performed based on the behavioral causal database to establish a predecessor path set; the predecessor path set is retrospectively verified to establish a retrospective verification trust level; and retrospective screening is performed using the retrospective verification trust level to complete the real abnormality retrospective analysis.

[0082] Furthermore, the platform is also used to implement the following functions: An unsupervised anomaly recognition model is established for each workstation; the multi-workstation monitoring data sets are respectively input into the unsupervised anomaly recognition model, and the workstation anomaly is output.

[0083] Furthermore, the platform is also used to implement the following functions: Historically normal artifacts are clustered to establish representative trajectory groups; time window sliding updates of the representative trajectory groups are performed based on a trajectory evolution mechanism; adjustment nodes of production parameters are extracted, confidence weights are established based on the adjustment nodes, and the confidence weights are used to compensate for the time window sliding updates to establish the healthy connector trajectory.

[0084] Furthermore, the platform is also used to implement the following functions: Record quality anomalies and generate workstation feature attention for quality anomalies, wherein the workstation feature attention is a time window attention, and the window size of the time window is positively correlated with the quality anomaly value; utilize the workstation feature attention to perform workstation monitoring and management for terminal connector production.

[0085] Furthermore, the platform is also used to implement the following functions: The abnormality time evolution prediction is performed using the backtracking positioning result and the workstation abnormality to establish an auxiliary authentication abnormality; and the terminal connector quality inspection result is compensated based on the auxiliary authentication abnormality.

[0086] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0088] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A terminal connector multi-station data collection and analysis method, characterized in that: The method comprises: Establishing a unified processing timeline for terminal connectors, tracking terminal connector production based on a unique binding ID based on the unified processing timeline, and establishing a multi-station monitoring data set; Performing workstation-based status anomaly identification on the multi-workstation monitoring data set to establish workstation anomalies; configuring a dynamic state chain using the multi-station monitoring data set and the processing timeline; Perform residual analysis based on the dynamic state chain and healthy connector trajectory to establish anomaly focus points; Based on the behavioral causal database between production process configuration workstations, a real anomaly retrospective analysis is performed according to the behavioral causal database and the anomaly focus point to establish a retrospective positioning result; A terminal connector quality inspection result is generated according to the backtracking positioning result and the workstation abnormality.

2. A terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: The performing residual analysis based on the dynamic state chain and the healthy connector trajectory to establish an abnormal focus point includes: Inputting the dynamic state chain and healthy connector trajectory into the twin residual channel; The representation extractor in the twin residual channel is used to extract features of time series, images, and numerical source data respectively, and perform feature fusion to output low-dimensional embedding vector pairs; The residual calculation layer of the twin residual channel performs residual calculation on the low-dimensional embedding vector pair and outputs anomaly scores through the discriminator; Anomaly focus points are established based on the anomaly scores.

3. A terminal connector multi-station data collection and analysis method according to claim 2, characterized in that: The establishing of an abnormal focus point according to the abnormality score includes: After normalizing the anomaly score, the normalized anomaly score is mapped to a workstation; After threshold screening of the normalized anomaly scores, focus analysis of adjacent workstations and time-continuous anomalies is performed based on the workstation mapping to construct focused segments. Abnormal focus points are generated using the focus segments.

4. The terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: The behavioral causal database between workstations configured based on the production process includes: Perform process analysis based on the production process and extract station attributes, wherein the station attributes include input and output materials, control parameters, and quality indicators; Performing a workstation dependency analysis based on the workstation attributes to establish first causal association data; calling an expert rule base according to the production process, performing rule item analysis of mapping process deviations and quality results according to the expert rule base, and establishing second causal association data; The behavior causal database is constructed based on the first causal association data and the second causal association data.

5. The terminal connector multi-station data collection and analysis method according to claim 4, characterized in that: The performing of real anomaly retrospective analysis based on the behavior causal database and the anomaly focus includes: Perform a predecessor path query based on the behavior causal database according to the abnormal focus point to establish a predecessor path set; Performing backtracking verification on the predecessor path set to establish backtracking verification trust; The backtracking verification trust level is used to perform backtracking screening to complete the real abnormality backtracking analysis.

6. The terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: The step of performing workstation-based status anomaly identification on the multi-workstation monitoring data set to establish workstation anomalies includes: Establish an unsupervised anomaly recognition model for each workstation; The multi-station monitoring data sets are respectively input into the unsupervised anomaly recognition model to output the workstation anomalies.

7. The terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: Before performing residual analysis based on the dynamic state chain and the healthy connector trajectory, the method includes: Cluster historical normal artifacts to establish representative trajectory groups; Performing a sliding update of the time window representing the trajectory group based on a trajectory evolution mechanism; Adjustment nodes of production parameters are extracted, confidence weights are established according to the adjustment nodes, and the confidence weights are used to compensate for time window sliding updates to establish the healthy connector trajectory.

8. The terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: Generating a terminal connector quality inspection result according to the backtracking positioning result and the workstation abnormality includes: Recording quality anomalies and generating workstation feature attention for quality anomalies, wherein the workstation feature attention is a time window attention, and the window size of the time window is positively correlated with the quality anomaly value; The workstation features are utilized to focus on workstation monitoring and management of terminal connector production.

9. The terminal connector multi-station data collection and analysis method according to claim 1, characterized in that: The generating of the terminal connector quality inspection result according to the backtracking positioning result and the workstation abnormality also includes: Using the backtracking positioning results and the workstation anomaly to predict the anomaly time evolution, and establish auxiliary authentication anomaly; The terminal connector quality inspection result is compensated based on the auxiliary authentication abnormality.

10. A terminal connector multi-station data acquisition and analysis platform, characterized in that: The platform is used to implement the terminal connector multi-station data acquisition and analysis method according to any one of claims 1 to 9, and the platform includes: A monitoring data set establishment module is used to establish a unified processing timeline for terminal connectors, perform terminal connector production tracking based on a unique binding ID based on the unified processing timeline, and establish a multi-station monitoring data set; A workstation anomaly establishment module is used to identify state anomalies based on the workstations of the multi-workstation monitoring data set and establish workstation anomalies; A dynamic state chain configuration module, configured to configure a dynamic state chain using the multi-station monitoring data set and the processing time axis; An abnormal focus point establishment module, used for performing residual analysis based on the dynamic state chain and the healthy connector trajectory to establish an abnormal focus point; A backtracking location result establishment module is used to configure a behavior causal database between workstations based on the production process, perform real anomaly backtracking analysis based on the behavior causal database and the anomaly focus point, and establish a backtracking location result; The quality inspection result generating module is used to generate the terminal connector quality inspection result according to the backtracking positioning result and the workstation abnormality.

Images