Automobile injection mold forming defect associated data processing method and system

Abstract

The invention provides an automobile injection mold forming defect associated data processing method and system, and relates to the technical field of defect prediction.The method comprises the steps that 2, the length of a time window is set, a standardized event item set and a pre-recorded forming defect type are aligned and spliced according to the time window, and a forming defect transaction data set is constructed; and step 3, for the forming defect transaction data set, screening a frequent item set meeting the minimum support degree by counting the occurrence frequency of each event item set in the transaction, and extracting an association rule between a condition event formed by an alarm event item and a material replacement event item and the forming defect based on the weighted support degree and a confidence coefficient threshold value, and forming a defect association rule set. According to the invention, prediction and real-time early warning of the molding defects of the automobile injection mold are realized, and the defect identification accuracy and the production quality management and control efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of defect prediction technology, and in particular to a method and system for processing data related to molding defects in automotive injection molds. Background Technology

[0002] In the process of automotive injection molding, the prediction and control of molding defects is a key link in ensuring product quality. Most existing defect analysis methods rely on the mining of association rules from historical production data. By identifying the statistical relationship between frequently occurring alarm events or material change events and the final defects, a static rule base is established for online quality early warning.

[0003] However, in practical applications, these static rule-based prediction methods often face limitations. For example, in a car bumper injection molding production scenario, process engineers found that when the mold experiences abnormal temperature, triggers an alarm, and is accompanied by a batch of masterbatch replacement events, the static rule library triggers a high probability of predicting shrinkage defects. However, in actual production, the same alarm event and material replacement combination sometimes leads to severe shrinkage, while at other times it has little impact on product quality. After investigation, it was found that this difference is often related to the real-time fluctuation of process parameters, especially the small torsion angle changes during screw rotation. This change reflects the uniformity and stability of the melt during the plasticization stage. However, existing rule matching mechanisms usually only focus on whether the event occurs, making it difficult to incorporate such microsecond-level process parameter fluctuation characteristics into the prediction model, resulting in the inability to maintain stable accuracy of prediction results under dynamic operating conditions. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for processing data related to molding defects in automotive injection molds, so as to realize the prediction and real-time early warning of molding defects in automotive injection molds, and improve the accuracy of defect identification and the efficiency of production quality control.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A first aspect is a method for processing data related to molding defects in automotive injection molds, the method comprising: Step 1: Collect multi-source process data of automotive injection mold during the molding process, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set; Step 2: Set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and construct the forming defect transaction dataset; Step 3: For the molding defect transaction dataset, the frequency of each event item set in the transaction is statistically analyzed to filter the frequent itemsets that meet the minimum support. Based on the weighted support and confidence threshold, the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects are extracted to form a defect association rule set. Step 4: Based on the defect association rule set, perform rule matching on the online process data stream to calculate the initial prediction probability of forming defects; extract the real-time monitoring values ​​of three key process parameters from the online process data stream, calculate the screw torsion angle parameter by solving the screw torque and speed data, construct process characteristic indicators and extract local fluctuation features to obtain process stability features, generate dynamic correction coefficients to correct the initial prediction probability, and obtain the final prediction probability to generate the corresponding early warning signal.

[0006] Secondly, the automotive injection mold molding defect correlation data processing system includes: The data acquisition module is used to collect multi-source process data during the molding process of automotive injection molds, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set. The building module is used to set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and build the forming defect transaction dataset; The filtering module is used to filter frequent itemsets that meet the minimum support by statistically analyzing the frequency of each event itemset in the transaction dataset of molding defects. Based on the weighted support and confidence threshold, it extracts the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects, forming a defect association rule set. The matching module is used to perform rule matching on the online process data stream based on the defect association rule set, calculate the initial prediction probability of forming defects, extract the real-time monitoring values ​​of three key process parameters from the online process data stream, calculate the screw torsion angle parameter by solving the screw torque and speed data, construct process characteristic indicators and extract local fluctuation features to obtain process stability features, generate dynamic correction coefficients to correct the initial prediction probability, obtain the final prediction probability and generate corresponding early warning signals.

[0007] The above-described solution of the present invention has at least the following beneficial effects: By converting continuous process parameters into discrete state event labels and uniformly encoding discrete alarm events and material change events into Boolean event items, standardized integration of different types of process data is achieved. By setting the time window length and aligning the standardized event item set with the molding defect type according to the time window, the temporal correspondence between process events and defect results is effectively established, improving the accuracy of correlation analysis. Based on weighted support and confidence thresholds, the association rules between conditional events composed of alarm event items and material change event items and molding defects are extracted, revealing the key process event combinations that lead to defects and providing a clear direction for process optimization. By extracting real-time monitoring values ​​of key process parameters, solving screw torsion angle parameters to construct process characteristic indicators, and extracting local fluctuation features to obtain process stability characteristics, dynamic correction coefficients are generated to correct the initial prediction probability, improving the accuracy of defect early warning. Based on online process data stream, rule matching and probability calculation are performed to generate early warning signals and reduce the generation of defective products. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the method for processing data related to molding defects in automotive injection molds provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of the automotive injection mold molding defect association data processing system provided in an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown, embodiments of the present invention propose a method for processing data related to molding defects in automotive injection molds, the method comprising the following steps: Step 1: Collect multi-source process data of automotive injection mold during the molding process, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set; Step 2: Set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and construct the forming defect transaction dataset; Step 3: For the molding defect transaction dataset, the frequency of each event item set in the transaction is statistically analyzed to filter the frequent itemsets that meet the minimum support. Based on the weighted support and confidence threshold, the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects are extracted to form a defect association rule set. Step 4: Based on the defect association rule set, perform rule matching on the online process data stream to calculate the initial prediction probability of forming defects; extract the real-time monitoring values ​​of three key process parameters from the online process data stream, calculate the screw torsion angle parameter by solving the screw torque and speed data, construct process characteristic indicators and extract local fluctuation features to obtain process stability features, generate dynamic correction coefficients to correct the initial prediction probability, and obtain the final prediction probability to generate the corresponding early warning signal.

[0012] In this embodiment of the invention, by converting continuous process parameters into discrete state event labels and uniformly encoding discrete alarm events and material change events into Boolean event items, the standardized integration of different types of process data is achieved. By setting the time window length and aligning the standardized event item set with the molding defect type according to the time window, the temporal correspondence between process events and defect results is effectively established, improving the accuracy of correlation analysis. Based on weighted support and confidence thresholds, the association rules between the conditional events composed of alarm event items and material change event items and molding defects are extracted, revealing the key process event combinations that lead to defects and providing a clear direction for process optimization. By extracting real-time monitoring values ​​of key process parameters, calculating screw torsion angle parameters to construct process characteristic indicators, and extracting local fluctuation features to obtain process stability characteristics, dynamic correction coefficients are generated to correct the initial prediction probability, improving the accuracy of defect warning. Based on online process data streams, rule matching and probability calculation are performed to generate warning signals and reduce the generation of defective products.

[0013] In a preferred embodiment of the present invention, step 1 above involves collecting multi-source process data during the molding process of an automotive injection mold, converting continuous process parameters into discrete state event labels, and uniformly encoding discrete alarm events and material change events into Boolean event items to generate a standardized event item set, which may include: In an embodiment of the present invention, in step 110, multi-source process data generated by an automotive injection mold within at least one complete molding cycle is synchronously acquired to obtain an original multi-source data set; specifically including: First, determine the defined range of a complete molding cycle of the automotive injection mold. A complete molding cycle refers to the entire process from the mold closing of the injection molding machine, through melt injection, pressure holding, cooling and solidification, then to mold opening, product ejection, until the injection molding machine closes the mold again to prepare for the next injection, ensuring that the acquired data covers all stages of a complete production and avoiding subsequent processing deviations caused by data loss; Subsequently, through various data acquisition devices supporting the automotive injection production line, all multi-source process data generated within this complete molding cycle is synchronously acquired. During the acquisition process, ensure that the time stamps of all data are synchronized, that is, various types of data acquired at the same moment correspond to the same time mark, avoiding time misalignment of different types of data. Among them, the multi-source process data specifically includes three types of core data. The first type is continuous process parameter data obtained by sensors installed at key parts of the injection molding machine in real time. For example, the melt temperature data monitored by temperature sensors installed on each section of the barrel, the injection pressure data monitored by pressure sensors installed on the injection cylinder, the screw rotation speed data monitored by rotation speed sensors installed on the screw drive mechanism, the screw torque data monitored by sensors installed at the torque detection part, the holding pressure time data monitored by sensors installed in the holding pressure circuit, etc.; The second type is discrete equipment alarm event data automatically recorded by the injection molding machine equipment controller. For example, relevant data of various equipment abnormal alarms such as mold temperature abnormal alarm, injection pressure exceeding the standard alarm, screw rotation speed abnormal alarm, cooling system failure alarm, etc., including the time when the alarm occurs, the specific type of the alarm, the duration of the alarm, etc.; The third type is discrete material replacement record data manually input by production management personnel through the production management system or automatically recorded by the system. For example, relevant data of various material replacements such as color masterbatch replacement, raw material batch replacement, mold release agent replacement, etc., including the time of material replacement, the type and batch of the material before replacement, the type and batch of the material after replacement, the operator who performed the replacement operation, etc.; All the above synchronously acquired continuous process parameter data, discrete equipment alarm event data, and discrete material replacement record data are summarized to form an original multi-source data set.

[0014] Step 111: Extract continuous process parameter data from the original multi-source dataset. Use a pre-defined discretization rule to convert the continuous parameter value at each moment into a corresponding discrete state label, obtaining a discretized representation of the continuous parameters. Specifically, this includes: First, filtering and extracting all continuous process parameter data from the original multi-source dataset obtained in Step 110. During extraction, strictly distinguish between continuous and discrete data, extracting only continuously variable process parameters such as melt temperature, injection pressure, screw speed, screw torque, and holding time, excluding discrete equipment alarm event data and material change record data. After extraction, each extracted continuous process parameter data is individually organized and arranged according to the chronological order of timestamps, forming time series data corresponding to each continuous process parameter. This ensures that each moment has a corresponding process parameter value. If data is missing at certain moments, it is supplemented by averaging the parameter values ​​of two adjacent moments, ensuring the continuity and completeness of the continuous process parameter time series data. Subsequently, a unified discretization rule is pre-defined, which is set individually for each continuous process parameter to ensure that the discretization result accurately reflects the actual process parameter. In the process of setting up the variable state, the process combines the technological requirements of automotive injection molding, product quality standards, and historical production experience to determine the division range and corresponding discrete state label for each continuous process parameter. The specific discretization process is as follows: For each continuous process parameter, first determine the normal operating range of the parameter; then, based on the allowable range of process fluctuations, divide the normal operating range into multiple continuous numerical intervals. Simultaneously, values ​​exceeding the normal operating range are divided into upper limit exceedance intervals and lower limit exceedance intervals. Each numerical interval corresponds to a unique discrete state label, which uses easily distinguishable text labels. For example, the preset normal operating range for melt temperature parameters is 180℃-220℃, which is divided into four normal intervals: 180℃-190℃, 190℃-200℃, 200℃-210℃, and 210℃-220℃. These intervals correspond to four discrete state labels: melt temperature too low, melt temperature slightly low, melt temperature slightly high, and melt temperature too high. Additionally, intervals below 180℃ correspond to a melt temperature too low alarm label, and intervals above 220℃ correspond to a melt temperature too high alarm label. Similarly, the preset normal operating range and interval divisions for injection pressure parameters correspond to different discrete state labels.After the discretization rules are preset, for each continuous process parameter's time series data at each moment, the preset value interval to which the value belongs is determined, the corresponding discrete state label is found, and the continuous parameter value at that moment is replaced with the corresponding discrete state label. This process is repeated for all moments and all continuous process parameter values, ultimately yielding the discretized representation of each continuous process parameter, which is a sequence of discrete state labels arranged in chronological order.

[0015] Step 112: Define each discrete state label as an independent event item. Simultaneously, extract discrete equipment alarm event data and discrete material change record data from the original multi-source dataset. Define each alarm type and each material change behavior as an independent event item, resulting in an initial event item set composed of all event items. Specifically, this includes: First, for each discrete representation of a continuous process parameter obtained in Step 111, define each discrete state label contained therein as an independent event item. Each discrete state label corresponds to a unique event item, without repetition or omission. For example, each discrete state label such as low melt temperature, high melt temperature alarm, moderate injection pressure, and abnormal screw speed is treated as an independent event item, ensuring that each state change of a continuous process parameter corresponds to an independent event item. Then, from the original multi-source dataset obtained in Step 110, filter and extract all discrete equipment alarm event data and discrete material change record data. During the extraction process, retain the core information of both types of data. For equipment alarm event data, focus on extracting alarm type information, ignoring auxiliary information such as alarm duration and alarm trigger count. Alarms of the same type correspond to an independent event item, such as abnormal mold temperature alarm, excessive injection pressure alarm, cooling system failure alarm, screw drive abnormality alarm, etc. Each alarm type is defined as an independent event item. If there are multiple alarms of the same type, only one event item is defined for that alarm type, and no duplicate definition is made. Material change record data focuses on extracting material change behavior information. Each material change behavior corresponds to an independent event item, regardless of whether the material type being changed is the same or whether it is from the same batch. Each change operation is defined as an independent event item, such as the first color masterbatch change, the first raw material batch change, the second color masterbatch change, the release agent change, etc. Each change behavior is treated as an independent event item to ensure that each material change can be accurately represented as an event item. Finally, all the event items defined above are summarized and integrated, including discrete status label event items corresponding to continuous process parameters, discrete equipment alarm type event items, and discrete material change behavior event items. During the summarization process, duplicate event items are checked. If there are completely identical event items, the duplicates are deleted to ensure that all event items are unique, and finally, an initial event item set consisting of all independent event items is formed.

[0016] Step 113 assigns a unique event identifier to each independent event item in the initial event item set and encodes it as a Boolean variable. This ensures that for any given time, if the event item occurs or exists, the Boolean value is set to 1; otherwise, it is set to 0, resulting in a Boolean event item record with a timestamp. Specifically, this involves: first, systematically reviewing the initial event item set obtained in Step 112, assigning a unique event identifier to each independent event item in the set. The event identifier uses a numerical sequence or a combination of letters to ensure that the event identifier for each event item is unique and not repeated. For example, the event item with low melt temperature is assigned identifier 001, the event item with abnormal mold temperature alarm is assigned identifier 002, and the event item with the first color masterbatch replacement is assigned identifier 002. Item assignment identifier 003 is used to sequentially complete the identifier assignment for all event items. After the assignment is completed, a correspondence table between event items and event identifiers is established for easy querying and verification. Subsequently, each independent event item in the initial event item set is uniformly encoded as a Boolean variable. The Boolean variable only contains two values: 1 and 0. The encoding rule is uniformly set as follows: for each event item, combined with the time information in the original multi-source data collected in step 110, it is mapped to each specific moment to determine whether the event item has occurred or exists at that moment. If the event item has occurred or exists at that moment, the Boolean variable value corresponding to the event item is set to 1; if the event item has not occurred and does not exist at that moment, the Boolean variable value corresponding to the event item is set to 0.The specific judgment process is as follows: For discrete state label event items corresponding to continuous process parameters, determine whether the discretized representation of the continuous process parameter at this moment is the discrete state label corresponding to the event item. If yes, the Boolean value is set to 1; otherwise, it is set to 0. For example, if the discrete state label of melt temperature at a certain moment is "melt temperature is too low," then the Boolean value corresponding to this event item is set to 1, and the Boolean value of other event items corresponding to continuous process parameters is set to 0. For equipment alarm type event items, determine whether this type of equipment alarm has occurred at this moment. If yes, the Boolean value is set to 1; otherwise, it is set to 0. For example, if a mold temperature abnormality alarm occurs at a certain moment, then the Boolean value corresponding to the mold temperature abnormality alarm event item is set to 1, and the Boolean value of other alarm type event items is set to 0. For material change behavior event items, determine whether the material change behavior at this moment is... If the material change operation is executed or if the material change operation is valid at that moment, the Boolean value is set to 1; otherwise, it is set to 0. For example, if the first color masterbatch change operation is being executed at a certain moment, the Boolean value corresponding to the first color masterbatch change event item is set to 1, and the Boolean values ​​of other material change event items are set to 0. After completing the Boolean encoding for each moment and each event item, a corresponding timestamp is added to each set of encoding results. The timestamp is consistent with the timestamp of the original data collected in step 110 to ensure that each set of Boolean encoding results can correspond to a specific production moment, and finally form a Boolean event item record with timestamp. Each record contains a timestamp, the event identifier corresponding to all event items, and the corresponding Boolean value, which fully represents the occurrence or existence status of all event items at that moment.

[0017] Step 114: Based on the timestamped Boolean event item records, arrange them in chronological order and perform data cleaning on the generated event item sequence to finally generate a standardized event item set. Specifically, this includes: First, collecting all timestamped Boolean event item records obtained in Step 113, and sorting and organizing all records according to the order of the timestamps. The sorting process strictly follows the chronological logic, starting from the start time of the complete molding cycle of the automotive injection mold and arranging them sequentially to the end time of that molding cycle. This ensures that the event item sequence can completely and coherently reflect the dynamic changes of all event items throughout the entire molding cycle, avoiding problems such as disordered chronological order or missing records. After sorting, a time-ordered set is formed. An event item sequence is generated, where each record corresponds to a specific moment in time. All records are sequential and complete, covering the entire data collection period. Subsequently, this event item sequence undergoes data cleaning. The purpose of data cleaning is to remove invalid data and correct abnormal data, ensuring the accuracy and completeness of the event item sequence and providing a reliable data foundation for constructing the transaction dataset. The cleaning process includes three aspects: First, invalid records are removed. Invalid records include those with duplicate timestamps, timestamps exceeding the preset complete data collection period range, and records with missing Boolean values. For records with duplicate timestamps, only the earliest occurrence is retained, and all other duplicate records are removed. Records with timestamps exceeding the data collection period range are excluded because they do not belong to the target data collection period. According to the data collected in step 110, all records are removed. Records with missing Boolean values, i.e., records where the Boolean value corresponding to a certain event item is not marked as 1 or 0, cannot accurately represent the occurrence status of the event item and are all removed. Second, abnormal records are corrected. Abnormal records include records with incorrect Boolean value markings and records where the event identifier and Boolean value do not match. For example, if a mold temperature abnormality alarm clearly occurs at a certain moment, but the Boolean value corresponding to the event item is marked as 0, the Boolean value is corrected to 1 by checking the original equipment alarm data collected in step 110. For records where the event identifier and Boolean value do not match, such as matching the identifier of the melt temperature high event item with the Boolean value of the melt temperature low event item, the correspondence between the event item and the event identifier is checked. First, correct the correspondence to the correct one; second, supplement missing records. For cases where individual time points are missing after sorting, supplement the missing records by finding the records of the two adjacent time points before and after the missing time point, referring to the Boolean value status of each event item at the adjacent time point, and combining the corresponding process parameters, alarms, and material change information in the original multi-source data, to ensure the continuity of the event item sequence. After data cleaning, a complete, accurate, and coherent event item sequence is obtained. This sequence is arranged in chronological order and includes the Boolean value status of all event items at each time point in the complete molding cycle and the corresponding timestamp. This event item sequence is defined as a standardized event item set, which has a unified format and unified coding rules.

[0018] By comprehensively collecting and standardizing multi-source process data in the automotive injection molding process, the standardized integration of multiple types of process data has been achieved. Various discrete status labels, alarm types, and material change behaviors are defined as independent event items, and the correlation between various events and the molding process is determined, ensuring that every key event affecting molding quality can be accurately characterized.

[0019] In a preferred embodiment of the present invention, step 2 above, which involves setting a time window length and aligning and concatenating the standardized event itemset with the pre-recorded molding defect types according to the time window to construct a molding defect transaction dataset, may include: In this embodiment of the invention, step 220 involves, based on the time window length, taking the occurrence time of each pre-recorded molding defect as a benchmark, extracting all event items within the time window length from the standardized event item set to obtain the defect window event item set corresponding to each defect. Specifically, this includes: first, determining and setting the time window length, which is the core benchmark for subsequent event item extraction and transaction dataset construction. The setting process combines the actual molding cycle time of the automotive injection mold, the formation lag time of the molding defect, and historical production experience to ensure that the event items within the extracted time window comprehensively cover all relevant events that may lead to the molding defect, without omitting key influencing events or including long-term events unrelated to the defect. The molding cycle time, as defined in step 110, is the entire process from the moment the injection molding machine closes the mold, through melt injection, pressure holding, cooling and shaping, to mold opening, product ejection, and until the injection molding machine closes the mold again to prepare for the next injection. The time window length must be less than or equal to the duration of a single complete molding cycle to avoid limiting the extraction range. Excessive time can lead to interference from irrelevant events, while the time must be greater than the lag time of defect formation to ensure that all inducing events before the defect occurs can be captured. Subsequently, all pre-recorded molding defect information is retrieved. The pre-recorded molding defect information comes from the actual inspection records in the automotive injection molding process. Each defect record contains two core pieces of information: first, the specific time of occurrence of the molding defect, which is accurate to the same level as the Boolean event item record with timestamps in step 113 to ensure consistency in the time dimension; second, the specific type of molding defect, including various common defects in automotive injection molding processes such as shrinkage marks, material shortages, deformation, and bubbles, which are completely consistent with the defect types involved in subsequent steps 442 and 449. Next, each pre-recorded molding defect is processed one by one. For a single molding defect, its specific time of occurrence is first determined, and this time of occurrence is used as the benchmark point for time interception, that is, the end time of the time window. According to the preset time window length, all standardized event items within the corresponding time period are intercepted from the set of standardized event items generated in step 114. The interception process strictly follows the chronological logic, starting from the moment the defect occurs and tracing back to include the standardized event items corresponding to the moment the defect occurs, until the interception duration reaches the preset time window length. This ensures that all intercepted event items fall within the time window range, without omitting any event items within the window or including any event items outside the window range.

[0020] For example, if the preset time window length is 20 minutes, and a molding defect occurs at 14:30, and this time falls within the time range of the standardized event itemset, then 14:30 is taken as the end time of the time window. All standardized event items within the 20 minutes from 14:10 to 14:30 are extracted. All standardized event items (including event identifier, Boolean value status, and timestamp) corresponding to each time point are included in the extraction range, without omitting any time. After extraction, all standardized event items within the time window are summarized and integrated, retaining the complete information of each event item (event identifier, Boolean value status, and timestamp) to form the defect window event itemset corresponding to this molding defect. Following the same method, each pre-recorded molding defect is processed sequentially, and each defect generates an independent defect window event itemset, ensuring that each defect corresponds to all relevant event items within the time window preceding its occurrence, without omitting any defect or generating duplicate defect window event itemsets.

[0021] Step 221 involves associating each defect window event item set with its corresponding molding defect type to form a defect window transaction with a defect type label. Specifically, this includes: First, reviewing all defect window event item sets generated in Step 220, determining the defect record number corresponding to each defect window event item set, and simultaneously retrieving pre-recorded molding defect information to establish a one-to-one correspondence between defect record numbers and molding defect types. This ensures that each defect record number accurately corresponds to a specific molding defect type, without confusion or mismatch. Then, processing each defect window event item set individually, first finding the corresponding molding defect type based on its labeled defect record number, and using this molding defect type as the defect type label for that defect window event item set. The label content must be completely consistent with the pre-recorded defect type description, such as shrinkage mark or missing material, without any simplification or modification. Next, associating and integrating the defect window event item set with its corresponding defect type label. The association process must ensure that the correspondence between the two is unique and accurate, i.e., one defect window event... Each itemset corresponds to only one defect type label, and one defect type label can correspond to multiple defect window event itemsets (when multiple defects are of the same type). After association, a defect window transaction with a defect type label is formed. Each defect window transaction contains two core parts: first, the defect window event itemset corresponding to the defect (containing complete information of standardized event items at all times within the window); second, the defect type label corresponding to the defect (clearly indicating the defect type corresponding to the transaction). During the association process, the correspondence between each defect window event itemset and the defect type label must be checked one by one to avoid mismatches. For example, the defect window event itemset corresponding to a shrinkage defect may be mistakenly associated with a deformed defect type label. After the check is completed, all defect window transactions with defect type labels are organized and arranged in the order of defect record numbers. At the same time, the association check information of each transaction is recorded. The above operation is repeated until all defect window event itemsets are associated with their corresponding defect type labels, forming all defect window transactions. Each transaction completely contains both the defect window event itemset and the defect type label.

[0022] Step 222: Based on the time window length, extract all event itemsets within the time window that do not include the occurrence time of any forming defects from the standardized event itemset, and uniformly mark these event itemsets as defect-free type to obtain defect-free window transactions. Specifically, this includes: first, determining the extraction rules; the extracted time window must be exactly the same as the time window length set in step 220 to ensure consistency in the time dimension and avoid deviations in subsequent transaction dataset construction due to inconsistent window lengths; the extracted time window must not contain any pre-recorded occurrence times of forming defects, meaning that there are no forming defects within the entire duration of the window. If a defect occurs, ensure that the extracted event itemset corresponds to a defect-free production condition. Then, retrieve the standardized event itemset generated in step 114, clarifying the entire time range covered by the standardized event itemset (from the start time of the first complete molding cycle to the end time of the last complete molding cycle). Simultaneously, retrieve the occurrence times of all pre-recorded molding defects, summarizing and organizing all defect occurrence times to form a defect occurrence time list, facilitating subsequent judgment of whether a time window contains a defect occurrence time. Next, according to the preset time window length, continuously truncate the entire time range covered by the standardized event itemset using a sliding truncation method. The process involves dynamic truncation, with a sliding step size set to one acquisition time (consistent with the sliding step size logic of the sliding time window in step 447). Starting from the beginning of the standardized event itemset, the first time window (from the beginning time to the beginning time plus the time window length) is truncated. Then, every acquisition time interval, the process slides forward one time to truncate the next time window, and so on, until the end of the standardized event itemset is reached. This ensures comprehensive and continuous truncation of the entire standardized event itemset's time range, without missing any truncationable time windows. For each truncated time window, it is determined whether the window contains any missing time windows. For any defect occurrence time in the defect occurrence time list, the judgment criteria are as follows: if any defect occurs within the entire range from the start time to the end time of the time window, then the time window is a defect-containing time window and is discarded without further extraction; if no defect occurs within the entire range from the start time to the end time of the time window, then the time window is a defect-free time window, which is retained, and all standardized event items within the window are extracted and integrated to form a defect-free window event item set, retaining the complete information of each event item (event identifier, Boolean value status, timestamp).

[0023] For example, if the preset time window length is 20 minutes, the time range of the standardized event itemset is from 14:00 to 15:00, and the defect occurs at 14:30, then the extracted time windows include 14:00-14:20, 14:01-14:21, ..., 14:40-15:00. Among these, time windows including 14:30 (14:10-14:30, 14:11-14:31, ..., 14:30-14:50) are discarded, while time windows not including 14:30 (14:00-14:20, 14:01-14:21, ..., 14:09-14:29, 14:51-15:00) are retained. The event itemsets within each retained window are extracted to form the defect-free window event itemsets. After all defect-free time window event itemsets are extracted, each defect-free window event itemset is uniformly marked as defect-free. The defect type is marked as the defect type label for the defect-free window event itemset, and its format is consistent with the defect type label of the defect window transaction in step 221, except that the label content is defect-free type. Then, each defect-free window event itemset and its corresponding defect-free type label are associated and integrated to form a defect-free window transaction. Each defect-free window transaction contains two core parts: one is the defect-free window event itemset (containing complete information of standardized event items at all times within the window); the other is the defect-free type label. Finally, all defect-free window transactions are sorted and arranged according to the order of the captured time windows. At the same time, it is checked whether there are duplicate defect-free window transactions (i.e., transactions with completely identical event itemsets and completely overlapping time windows). If duplicate transactions exist, only one is retained and the rest are removed to ensure the uniqueness and accuracy of all defect-free window transactions, thus completing the acquisition of defect-free window transactions.

[0024] Step 223: Merge all defective window transactions with all non-defective window transactions to construct a complete defective transaction dataset. Each transaction includes a set of event items and a defect category label. Specifically, this involves: First, retrieving all defective window transactions obtained in Step 221 and all non-defective window transactions obtained in Step 222, and performing a preliminary review of the two types of transactions to ensure that the formats of the two types of transactions are consistent and the content is complete. Each transaction contains a set of event items (defective window event item set or non-defective window event item set) and a defect type label (specific defect type or non-defect type). The format of the event item set is consistent with the format of the standardized event item set generated in Step 114. The description of defect type labels is standardized and unambiguous, completely consistent with the pre-recorded defect type and defect-free type labels. Subsequently, a merging operation is performed on the two types of transactions. The merging process involves summarizing and integrating all defect-window transactions and all defect-free window transactions into a single set, without omitting any defect-window or defect-free window transactions, while not adding any additional transactions or modifying any transaction content (including event itemsets and defect type labels), ensuring the integrity and accuracy of the merged transactions. After merging, all integrated transactions are uniformly organized. This organization process includes two aspects: firstly, unifying the order of transactions, which can be based on the end of the time window corresponding to the transaction. First, transactions are arranged in chronological order, regardless of whether they are defect-free or defect-free, ensuring the continuity of the transaction dataset. Second, the format of transactions is standardized, assigning a unique transaction identifier to each transaction. The transaction identifier uses either a numerical sequence or a combination of letters to ensure that each transaction can be uniquely identified. Simultaneously, the core content of each transaction (event itemset, defect type label, and transaction identifier) ​​is standardized and organized, clearly distinguishing between the event itemset and the defect type label. After this organization, a complete defect transaction dataset is constructed. The core feature of this dataset is that each transaction contains a set of event items and a defect category label, where the event itemset corresponds to a time window. All standardized event items within the data set (fully retaining event identifiers, Boolean values, and timestamps) are labeled with specific molding defect types (such as shrinkage marks or missing material) or no-defect types. The transaction dataset contains transactions corresponding to both defective and non-defective working conditions, comprehensively covering different working conditions in the automotive injection molding process. Finally, a comprehensive check is performed on the molding defect transaction dataset. The check includes whether all transactions are complete (no missing or omitted items), whether the correspondence between the event item set and the defect type label of each transaction is accurate (no mismatches), whether the transaction identifier is unique (no duplicates), and whether the format of the event item set is standardized (consistent with the standardized event item set).

[0025] By associating event items with molding defects, and based on a uniform time window length, this method effectively distinguishes between event combinations that lead to defects and event combinations that occur during normal production. This avoids misjudging routine events in normal production as defect-related events, improves the accuracy of association rule extraction, and reduces the risk of misjudgment.

[0026] In a preferred embodiment of the present invention, step 3 above involves statistically analyzing the frequency of each event item set in the molding defect transaction dataset, filtering frequent item sets that meet the minimum support, and extracting association rules between conditional events composed of alarm event items and material replacement event items and molding defects based on weighted support and confidence thresholds, thus forming a defect association rule set, which may include: In this embodiment of the invention, step 330 involves statistically analyzing the frequency of occurrence of each single event item in the molding defect transaction dataset, calculating the support of each single event item, and filtering all single event items with a support greater than or equal to a preset minimum support threshold as a frequent item set. Specifically, this includes: first, determining the composition of the molding defect transaction dataset, which contains several transactions, each transaction consisting of a set of event items and a defect category label, wherein the event items include status event items after discretization of continuous process parameters, alarm event items after Boolean encoding, and material replacement event items after Boolean encoding; second, extracting each transaction in the molding defect transaction dataset one by one, separating all single event items contained in each transaction, and then independently statistically analyzing each single event item, counting the total number of times the single event item appears in all transactions, which is the frequency of occurrence of the single event item. For example, if the molding defect transaction dataset has 1000 transactions, and the single event item "temperature abnormal alarm" appears in 300 transactions, then the frequency of occurrence of the single event item is 300. Next, the support of each individual event item is calculated. The support is calculated by dividing the frequency of a single event item by the total number of transactions in the molding defect transaction dataset. That is, the support of a single event item = the frequency of the single event item ÷ the total number of transactions. For example, the frequency of the temperature abnormality alarm event item is 300, and the total number of transactions is 1000. Then, the support of this event item = 300 ÷ 1000. The result is the support of this event item. Finally, a minimum support threshold is preset. This threshold is preset based on the actual automotive injection mold production scenario and the distribution of historical defect data. It is used to filter out event items with a sufficiently high frequency and statistical significance. The support calculated for each individual event item is compared with the preset minimum support threshold one by one. All individual event items with a support value greater than or equal to the preset minimum support threshold are retained. These retained individual event items together constitute a frequent item set. Individual event items with a support value less than the preset minimum support threshold are removed to ensure that the frequent item set only contains event items whose frequency meets the statistical requirements.

[0027] Step 331: Based on frequent item sets, candidate two-item sets are generated through pairwise combinations. The support of each candidate two-item set in the formed defect transaction dataset is calculated, and item sets with support greater than or equal to a preset minimum support threshold are retained as frequent two-item sets. Specifically, this includes: first, retrieving the frequent item sets obtained in step 330, comprehensively reviewing and verifying them to confirm that all event items in the frequent item sets are frequent event items with support greater than or equal to the preset minimum support threshold, with no non-frequent event items mixed in. Simultaneously, it is confirmed that there are no duplicate or abnormal event items in the frequent item sets, ensuring the accuracy and completeness of the frequent item sets and providing a reliable foundation for the generation of candidate two-item sets. After reviewing, all event items in the frequent item sets are arranged according to the order of their event identifiers to facilitate the orderly implementation of pairwise combination operations and avoid combination chaos. Subsequently, based on the arranged frequent item sets, candidate two-item sets are generated through pairwise combinations. The generation of candidate two-item sets follows clear combination rules, specifically, starting from the frequent item sets... Two distinct single event items are randomly selected and combined to form a set containing both event items; this set is the candidate binomial set. During the combination process, the principles of no repeated combination and no self-combination are strictly followed. That is, the order of combining two event items is irrelevant; if {event item A, event item B} has already been generated, then {event item B, event item A} will not be generated again to avoid duplicate combinations. Simultaneously, a single event item is not allowed to combine itself, such as {event item A, event item A}, as such combinations are meaningless and invalid, and are excluded. The specific combination process is as follows: First, select the first event item in the frequent one-item set, and then combine it with each subsequent distinct event item in the frequent one-item set one by one to generate the corresponding candidate binomial set. Next, select the second event item in the frequent one-item set, and combine it with each subsequent distinct event item one by one, without repeating the combination with the first event item. This process continues until all distinct event items in the frequent one-item set have been paired, ensuring no valid combinations are omitted.

[0028] After the candidate two-item sets are generated, a list of candidate two-item sets is compiled. All candidate two-item sets generated through pairwise combinations are collected into the list. Each candidate two-item set in the list is compared one by one to confirm that there are no duplicate combinations or invalid combinations, such as combinations of themselves or combinations of irrelevant event items. At the same time, it is confirmed that each candidate two-item set consists of two different single event items from frequent one-item sets, with no event items from infrequent one-item sets mixed in, ensuring the completeness and accuracy of the candidate two-item sets. After the compilation is completed, a unique identifier is assigned to each candidate two-item set to facilitate the support calculation and verification work. Next, the support of each candidate two-item set is calculated. The statistical and calculation logic of the support of candidate two-item sets is completely consistent with the statistical and calculation logic of the support of single event items in step 330. The only difference is that the object of the statistics and calculation changes from single event items to candidate two-item sets containing two event items. The specific calculation process is divided into two steps. The first step is to statistically analyze the support of each candidate two-item set. Frequency of occurrence is defined as the total number of times that two single event items in a candidate binomial set appear simultaneously in the same transaction. That is, both event items actually occur or exist within the time window corresponding to the transaction. The corresponding Boolean value is 1. During the statistics, for each candidate binomial set (selected one by one according to its unique identifier), each transaction in the defective transaction dataset is compared one by one to check whether the single event items extracted from the transaction contain both single event items from the candidate binomial set. If they are both contained, the candidate binomial set is determined to have appeared once in the transaction, and the count counter is incremented by 1. If only one event item is contained, or neither event item is contained, the candidate binomial set is determined not to have appeared in the transaction, and the counter remains unchanged. After comparing all transactions, the final value of the counter is the frequency of occurrence of the candidate binomial set. During the statistical process, omissions and errors in statistics are also avoided to ensure the accuracy of the frequency of occurrence statistics for each candidate binomial set.

[0029] The second step is to calculate the support of the candidate two-item sets. The calculation method is to divide the frequency of occurrence of the candidate two-item set by the total number of transactions in the defective transaction dataset. During the calculation, the order of frequency of occurrence of the candidate two-item set ÷ total number of transactions must be strictly followed, without changing the calculation logic, maintaining consistency with the calculation method for the support of a single event item. For example, if a candidate two-item set occurs 30 times and the total number of transactions in the defective transaction dataset is 200, then the support of the candidate two-item set is 30 divided by 200. This support represents the probability that both events occur simultaneously in the defective transaction dataset. The higher the support level, the greater the probability of these two events occurring simultaneously, and the closer their potential correlation with molding defects. After calculating the support level for each candidate binomial item set, the calculation results are linked to the identifier and frequency of occurrence of that candidate binomial item set for easy screening and verification. After the support level calculation is completed, the minimum support threshold preset in step 330 is used without changing any parameters of this threshold to ensure the consistency of the screening criteria and avoid deviations in the screening results due to threshold changes. Subsequently, frequent binomial items are screened according to the same rules as for single event items. The selection process involves comparing the support of each candidate two-item set with a preset minimum support threshold. If the support of the candidate two-item set is greater than or equal to the preset minimum support threshold, it is determined to be a frequent two-item set, indicating that the two event items occur frequently and are worthy of further analysis; therefore, it is selected. If the support of the candidate two-item set is less than the preset minimum support threshold, it is determined to be an infrequent two-item set, indicating that the two event items occur very infrequently and have a very low probability of being associated with molding defects; therefore, it is eliminated and not included in the processing. After the selection is completed, all... The frequent two-items sets that meet the requirements are aggregated to form a frequent two-items set set. The identifier, frequency of occurrence, and support information of each frequent two-items set in this set are retained to facilitate the generation of higher-order itemsets. After the aggregation is completed, each frequent two-items set in the set is checked one by one to confirm that its support is greater than or equal to the preset minimum support threshold, and that it is composed of two different event items from the frequent one-items set, without any anomalies or errors, to ensure the accuracy and reliability of the frequent two-items sets. Frequent two-items sets are mainly used to reflect the pattern of two high-frequency event items occurring simultaneously, providing a basis for generating candidate three-items sets and screening frequent three-items sets.

[0030] Step 332: Based on frequent two-item sets, candidate three-item sets are generated through a join operation. The support of each candidate three-item set is calculated, and itemsets with a support greater than or equal to a preset minimum support threshold are retained as frequent three-item sets. This process is repeated until no new candidate itemsets can be generated, resulting in all frequent itemsets. Specifically, this includes: first, based on the frequent two-item sets obtained in step 331, a join operation is performed to generate candidate three-item sets. The join operation specifically involves selecting two two-item sets from the frequent two-item sets that contain a common event item, merging these two two-item sets, removing the duplicate event item, and forming a set containing three different event items. This involves selecting candidate three-item sets. For example, frequent two-item sets might contain {temperature anomaly alarm, colorant replacement} and {temperature anomaly alarm, excessive pressure alarm}. Both sets share the same event item: temperature anomaly alarm. Merging them and removing duplicates yields the candidate three-item set {temperature anomaly alarm, colorant replacement, excessive pressure alarm}. If two sets in a frequent two-item set do not share any event items, no join operation is performed to avoid generating invalid candidate three-item sets. Next, the frequency of each candidate three-item set is counted, using the same method as for the candidate two-item set frequency count, i.e., checking each completed defect transaction dataset one by one. For each transaction in the dataset, it is determined whether the transaction simultaneously contains all three event items from a candidate three-item set. If so, the candidate three-item set is considered to have appeared once in the transaction. The total number of times the candidate three-item set appears across all transactions is the frequency of the candidate three-item set. Next, the support of each candidate three-item set is calculated using the formula: Candidate Three-Item Set Support = Frequency of the Candidate Three-Item Set ÷ Total Number of Transactions in the Defect Transaction Data Set. The support value for each candidate three-item set is then compared with a preset minimum support threshold. Candidate three-itemsets with support greater than or equal to the threshold are retained; these are frequent three-itemsets. Candidate three-itemsets with insufficient support are removed. Then, following the same logic, the process is repeated cyclically. Based on the frequent three-itemsets, candidate four-itemsets are generated through a join operation (selecting three-itemsets containing two identical event items and merging them to remove duplicates). The frequency of occurrence of the candidate four-itemsets is then counted, their support is calculated, and compared with the minimum support threshold to filter out the frequent four-itemsets. This process is repeated, with each step generating candidate itemsets for the current level based on the frequent itemsets of the previous level, and then filtering the frequent itemsets of the current level through frequency statistics, support calculation, and threshold comparison.Finally, repeat the above operation until no new candidate itemsets can be generated (i.e., after generating candidate itemsets of a certain order, if the support of all candidate itemsets is less than the preset minimum support threshold after frequency statistics and support calculations, it is impossible to filter out frequent itemsets of the corresponding order, or it is impossible to generate new candidate itemsets by connecting frequent itemsets of the previous order). At this point, stop the filtering operation, and summarize all the frequently obtained one-itemets, frequent two-itemets, frequent three-itemets, and higher-order frequent itemsets to obtain all frequent itemsets.

[0031] Step 333: For each frequent itemset, generate all possible association rules that satisfy the conditions. The condition event set consists of alarm event items and material change event items, and the defect type is the defect category label recorded in the transaction dataset. Specifically, this includes: First, extracting all frequent itemsets obtained in step 332, and performing rule generation operations for each frequent itemset. Each frequent itemset corresponds to several possible association rules. The overall form of the association rule is condition event set to defect type, where the left side of the arrow represents the condition event set, and the right side of the arrow represents the corresponding defect type. Second, clarifying the composition requirements of the association rules... The event set can only consist of alarm event items and material change event items, and cannot include state event items after discretization of continuous process parameters; the defect type must be a defect category label pre-recorded in the molding defect transaction dataset (such as shrinkage mark, crack, missing material, etc.), and this defect type must be consistent with the defect category label corresponding to the transaction in which the frequent itemset on which the generation rule is based; then, for each frequent itemset, all alarm event items and material change event items contained therein are extracted, and these alarm event items and material change event items are combined to form different condition event sets (which can be generated according to the order of the frequent itemset). Different numbers of conditional event sets are used to ensure that all reasonable combinations of alarm event items and material change event items are covered. Simultaneously, the defect type corresponding to the transaction containing the frequent itemset is determined, and each conditional event set is mapped to that defect type, generating association rules between conditional event sets and defect types. For example, if a frequent three-item set contains temperature anomaly alarm (alarm event item), color masterbatch change (material change event item), and melt temperature too high (status event item), and the defect type corresponding to the transaction containing this frequent itemset is shrinkage, then the alarm event items and material change event items in this frequent itemset are split to form conditional event sets. The item set can be {temperature anomaly alarm}, {color masterbatch replacement}, or {temperature anomaly alarm, color masterbatch replacement}, corresponding to the defect type of shrinkage mark, respectively. Three association rules are generated: temperature anomaly alarm to shrinkage mark, color masterbatch replacement to shrinkage mark, and temperature anomaly alarm + color masterbatch replacement to shrinkage mark. Conditional event sets containing the state event item of melt temperature too high are removed to ensure that all generated association rules meet the requirements of the conditional event set. Finally, the above operation is performed on each frequent itemset to generate all possible association rules that meet the conditions. After summarizing, an initial association rule set is obtained to ensure that no association rule that meets the requirements is missed.

[0032] Step 334: For each association rule, calculate the confidence level of each association rule, and assign corresponding weights to the event items involved in each rule based on the importance of the event, and calculate the weighted support of each rule. Specifically, this includes: First, retrieving all association rules generated in step 333, reviewing each association rule one by one, clarifying the condition event set (the combination of alarm event items and material replacement event items) and result event (defect category label) of each rule, ensuring that the condition event set and result event of each association rule are clear, accurate, unambiguous, and error-free. Then, for each association rule, calculate its confidence level and weighted support. The confidence level calculation process is as follows: Confidence level is used to characterize the probability of the result event (corresponding to the defect type) occurring when all event items in the condition event set occur simultaneously. The calculation method... The method involves first counting the number of transactions in the established defect transaction dataset that simultaneously contain the conditional event set and the result event of the association rule, i.e., the number of transactions in which the conditional event set occurs and the corresponding defect type appears. Then, it involves counting the number of transactions in the established defect transaction dataset that only contain the conditional event set of the association rule, i.e., the total number of transactions in which the conditional event set occurs. Regardless of the corresponding defect type, the confidence of the association rule is obtained by dividing the number of transactions that simultaneously contain the conditional event set and the result event by the number of transactions that only contain the conditional event set. For example, if the number of transactions that simultaneously contain the conditional event set and the result event is 30 and the number of transactions that only contain the conditional event set is 40, then the confidence of the association rule is 30 divided by 40. The higher the confidence, the stronger the association between the conditional event set and the corresponding defect type.

[0033] After the confidence level is calculated, the weighted support of each association rule is calculated. The weighted support calculation requires assigning corresponding weights to the event items involved in each rule. The weights are assigned based on the importance of the event item, which is determined by its impact on molding defects. Event items with a greater impact on molding defects are assigned higher weights, while those with a smaller impact are assigned lower weights. Alarm events and material change events have higher weights than discrete state label events corresponding to continuous parameters. However, this step only involves alarm and material change events in the association rules. Specifically, severe alarm events, such as severe abnormal mold temperature alarms and severe alarms exceeding the upper limit of injection pressure, have higher weights than general alarm events; critical materials, such as injection molding raw material change events, have higher weights than auxiliary materials, such as mold release agent change events.

[0034] After weighting is assigned, ensure that the sum of the weights of all event items in each association rule is 1. If the sum of the weights is not equal to 1, adjust the weights of all event items by dividing the current weight of each event item by the sum of the current weights of all event items. After adjustment, verify again to ensure that the sum of the weights of all event items is 1. After the weight adjustment, apply it to the weighted support calculation of the association rule. Then, calculate the weighted support of each association rule by extracting the weight of each event item in the conditional event set of the association rule and the support of each event item in the formed defect transaction dataset. Multiply the weight of each event item by its support to obtain the weighted support component of each event item. Then apply the weighted support component of the association rule conditional event set to the conditional event set of the association rule. The weighted support components of all event items are summed to obtain the weighted support of the association rule. For example, if the conditional event set of an association rule contains two event items, event item 1 has a weight of 0.6 and a support of 0.25, and event item 2 has a weight of 0.4 and a support of 0.3, then the weighted support component of event item 1 is 0.6 multiplied by 0.25, and the weighted support component of event item 2 is 0.4 multiplied by 0.3. The sum of these two components is the weighted support of the association rule. Each association rule is processed one by one, and the confidence and weighted support are calculated separately. During the calculation process, the above calculation logic is strictly followed, and each calculation step and each value is checked one by one to avoid calculation errors and ensure that the confidence and weighted support of each association rule are calculated accurately. The calculation results of each rule are also recorded.

[0035] Step 335: Compare the weighted support of each rule with a preset weighted support threshold, and simultaneously compare the confidence of each rule with a preset confidence threshold. Only retain association rules whose weighted support and confidence both exceed their respective thresholds, and compile them into a defect association rule set. Specifically, this includes: First, determining the preset weighted support threshold and the preset confidence threshold. The setting of both thresholds is based on production practice experience, the actual application requirements of association rules, and the distribution of weighted support and confidence of all association rules calculated in step 334. Fixed values ​​are not used, and the two thresholds are independent of each other and do not affect each other. The process involves two thresholds: a preset weighted support threshold to filter association rules with high overall value (frequent and important event items), and a preset confidence threshold to filter association rules with strong correlation (close association between the condition event set and the defect type). These thresholds ensure that redundant, invalid, and low-correlation association rules are eliminated, while key and reliable association rules are retained. After the thresholds are set, all association rules generated in step 334 are filtered one by one. The filtering rule is to compare the weighted support with the preset weighted support threshold and the confidence with the preset confidence threshold for each association rule. Only when the weighted support is higher than the preset weighted support threshold will the association rule be filtered out. An association rule is retained only if its weighted support is greater than or equal to a preset weighted support threshold and its confidence level is greater than or equal to a preset confidence threshold. If the weighted support of the association rule is less than the preset weighted support threshold, or its confidence level is less than the preset confidence threshold, or both are less than the preset confidence threshold, then the association rule is removed and not retained. During the screening process, the two calculated results (weighted support and confidence level) of each association rule are checked against the corresponding thresholds one by one. The screening rules are strictly followed without relaxing or raising the screening standards to avoid missing key rules or mistakenly retaining invalid rules. After the screening is completed, all retained association rules are... The rules are summarized. During the summary, each retained association rule is checked one by one to confirm that its weighted support and confidence levels meet the standards. The condition event set only includes alarm event items and material change event items. The result event is a valid defect category label with no anomalies or errors. After the summary is completed, the retained association rules are formatted and sorted. They are classified and organized according to the defect type corresponding to the association rule. For example, all association rules with result events of shrinkage defects are grouped into one category, and association rules with result events of no defect type are grouped into another category. After classification and organization, a complete set of association rules is formed, which is the defect association rule set.

[0036] It has enabled the discovery of the correlation between key events and molding defects. The generated defect correlation rule set can clearly and accurately reflect the correlation patterns between alarm events, material change events and various molding defects, promoting the reliability of the entire defect correlation data processing method and improving the molding quality control level of automotive injection molds.

[0037] In a preferred embodiment of the present invention, step 4 above, based on the defect association rule set, performs rule matching on the online process data stream to calculate the initial prediction probability of forming defects; extracts real-time monitoring values ​​of three key process parameters from the online process data stream; calculates the screw torsion angle parameter obtained by solving the screw torque and speed data; constructs process characteristic indicators and extracts local fluctuation features to obtain process stability features; generates dynamic correction coefficients to correct the initial prediction probability; and obtains the final prediction probability to generate a corresponding early warning signal, which may include: In this embodiment of the invention, step 440 involves real-time acquisition of online process data streams, standardizing events in the data streams, converting them into standardized event items, and forming an online event sequence. Specifically, this includes: first, starting the online data acquisition equipment of the automotive injection molding production line, ensuring real-time linkage between the acquisition equipment and the injection molding machine, sensors, and production management system to achieve continuous, real-time acquisition of the online process data stream. The acquisition frequency is consistent with the acquisition frequency of the original multi-source data in step 110, ensuring the timeliness and continuity of the data. During the acquisition process, the timestamp corresponding to each data item is recorded synchronously, with the timestamp accurate to the Boolean event item with timestamp in step 113. To ensure consistent accuracy and avoid time discrepancies affecting subsequent rule matching, the online process data stream content must be consistent with the original multi-source dataset collected in step 110. Specifically, it includes three core data categories: first, continuous process parameter data obtained in real-time from sensors installed at key parts of the injection molding machine, including melt temperature, injection pressure, holding time, screw torque, and screw speed; second, discrete equipment alarm event data automatically recorded by the injection molding machine controller, including the occurrence time and alarm type of various equipment anomaly alarms; and third, discrete material change record data manually entered by production management personnel through the production management system or automatically recorded by the system, including… This includes the time of material changeover, the material type and batch before and after the changeover, etc.; subsequently, all events in the real-time collected online process data stream are standardized according to the standardization rules set in steps 111 to 113, in accordance with the same standardization processing as in step 1, to ensure that the processed event items are in the same format and consistent with the standardized event item set generated in step 1. The specific processing process is as follows: for continuous process parameter data in the online process data stream, according to the discretization rules preset in step 111, the continuous parameter values ​​at each moment are converted into corresponding discrete state labels. The conversion process is completely consistent with step 111, that is, first determine the preset value range to which the parameter value belongs, and then... Each alarm type is assigned a unique discrete status label. For discrete equipment alarm event data in the online process data stream, each alarm type is treated as an independent event item according to the definition rules in step 112, ensuring complete consistency with the equipment alarm type event items defined in step 112. No new or modified event items are added or modified for each alarm type. For discrete material change record data in the online process data stream, each material change behavior is treated as an independent event item according to the definition rules in step 112, ensuring complete consistency with the material change behavior event items defined in step 112, thus ensuring that each online material change corresponds to a standardized event item.Next, following the encoding rules of step 113, a unique event identifier is assigned to each processed event item. This event identifier is completely consistent with the identifier assigned to the corresponding event item in step 113, ensuring the uniformity and identifiability of the event items. Subsequently, each event item is uniformly encoded as a Boolean variable, with the encoding rules consistent with step 113: at any given time, if the event item occurs or exists, the Boolean value is set to 1; otherwise, it is set to 0. Finally, all standardized and encoded event items are arranged in chronological order according to their timestamps. Each time point corresponds to a set of standardized event item Boolean values ​​and corresponding event identifiers, forming a continuous and coherent online event sequence. This online event sequence is completely consistent with the format of the standardized event item set generated in step 114, except that the data is real-time collected online data.

[0038] Step 441: Based on the time window length, using the current moment as a benchmark, extract all standardized event items falling within the time window from the online event sequence to form the current window event item set. Specifically, this includes: first, determining the time window length set in Step 2. This time window length is a fixed length preset when constructing the molding defect transaction dataset. Step 441 directly uses this time window length without any modification to ensure the uniformity of the time window and avoid deviations in rule matching due to inconsistent window lengths. The time window length is set based on the same criteria as in Step 2, combined with the molding cycle time of automotive injection molds, the lag time of defect formation, and historical production experience to ensure that the event items within the extracted time window can comprehensively cover all types of events that may lead to defects. Subsequently, determining the specific time point of the current moment, which is the latest moment of online data acquisition and also the benchmark moment for defect prediction. The current moment is updated synchronously in real time to ensure that each prediction is based on the latest online process data. Using the current time as the end time of the time window, and according to the preset time window length, all standardized event items within the corresponding duration are extracted. The extraction process strictly follows chronological logic, tracing back from the current time, including the current time, until the extracted duration reaches the preset time window length. This ensures that all extracted event items fall within the time window range, without omitting any event item within the window or including any event item outside the window. For example, if the preset time window length is 30 minutes and the current time is 10:00, then 10:00 is the end time, and all standardized event items within the 30 minutes from 9:30 to 10:00 are extracted, with all standardized event items corresponding to each time point included in the extraction range. After extraction, all standardized event items within the time window are summarized and integrated, retaining the Boolean value state, event identifier, and timestamp corresponding to each event item to form the current window event item set. This event item set fully represents the occurrence or existence state of all event items in the automotive injection mold forming process within the time window preceding the current time, providing specific event data support for rule matching.

[0039] Step 442: Obtain the defect association rule set. Each association rule includes antecedent condition event set, consequent defect type, and the corresponding confidence level. Specifically, this includes: first, determining that the defect association rule set is the rule set formed after filtering and extraction in step 3; then, directly obtaining this defect association rule set in step 442 without any modification, addition, or deletion to ensure the completeness and accuracy of the association rules and avoid deviations in prediction results due to rule tampering; during the acquisition process, ensuring that all rules in the defect association rule set are completely extracted, without omitting any association rule that meets the weighted support and confidence thresholds, while retaining all core information of each rule without missing any key content; the core components of each association rule are completely retained, specifically including three parts. The first part is the antecedent condition event set, which consists of alarm event items and material replacement event items defined in step 112. That is, the antecedent of each rule is a set composed of one or more equipment alarm type event items and material replacement behavior event items, and each event item corresponds to a unique event identifier assigned in step 113. The antecedent condition event set clarifies the combination of events required to trigger the rule; the second part is the consequent defect type, which is the molding defect type pre-recorded in step 2, including shrinkage marks, missing material, deformation, bubbles, and other common defects in automotive injection molding processes. Each rule corresponds to a specific defect type, clarifying the predicted defect type after the rule is triggered; the third part is the confidence level corresponding to the rule, which is the value calculated in step 3, reflecting the probability of the consequent defect type occurring when the antecedent condition event set occurs. The confidence level calculation process strictly follows the requirements of step 3, and the confidence level value is directly used in step 442 without recalculation; after obtaining the defect association rule set, the rules are sorted out one by one and initially classified according to the consequent defect type to facilitate the one-by-one comparison in step 443 and the group statistics in step 444. During the sorting process, only classification is performed without changing any content of the rules, ensuring that the antecedent condition event set, consequent defect type, and confidence level of each rule remain unchanged, ultimately forming a defect association rule set that can be directly used for rule matching.

[0040] Step 443: Compare the current window event item set with the antecedent condition event set of each rule in the defect association rule set one by one to determine whether the current window event item set completely includes all event items in the antecedent condition event set of the rule. If it does, the rule is determined to be triggered, and the triggered rule, the confidence level of the rule, and the consequent defect type of the rule are recorded to obtain the list of triggered rules for the current window. Specifically, this includes: First, taking out the current window event item set obtained in step 441, sorting out all event items contained in the event item set, clarifying the event identifier and corresponding Boolean value status of each event item, focusing on event items with a Boolean value of 1, that is, event items that have occurred or exist in the current window, and event items with a Boolean value of 0, indicating that they have not occurred and are not included in the subsequent comparison and judgment, ensuring that the comparison process focuses on the actual events that have occurred; At the same time, taking out the defect association rule set obtained in step 442, and sorting out the rules according to the sorted rules... The process proceeds sequentially, extracting each rule one by one and comparing them sequentially. The comparison is performed on a rule-by-rule basis, ensuring no rule is missed and no rule is compared repeatedly. Then, for each rule, the precondition event set is extracted, identifying the event identifiers of all event items included in the precondition event set. This includes all alarm events and material replacement events that must occur or exist simultaneously to trigger the rule. Each event item in the precondition event set is compared one by one with the event items in the current window event item set that have a Boolean value of 1. The determination is made whether the current window event item set contains all the event items from the precondition event set. The criterion is that each event item in the precondition event set has a corresponding event identifier in the current window event item set, and the corresponding Boolean value is 1, indicating that the event item has indeed occurred or exists within the current window.

[0041] The specific comparison process is as follows: Assume that the antecedent condition event set of a certain rule includes two event items: mold temperature abnormality alarm and first color masterbatch replacement, with corresponding event identifiers 002 and 003 respectively. Then, it is necessary to search for event items with event identifiers 002 and 003 in the current window's event item set and check if the corresponding Boolean values ​​of these two event items are both 1. If the Boolean values ​​of both event items are 1, it means that the current window's event item set completely includes all event items in the antecedent condition event set of this rule, and it is determined that the rule has been triggered. If the Boolean value of any event item is 0, or if there is no event item corresponding to that event identifier in the current window's event item set, it means that the current window's event item set does not completely include all event items in the antecedent condition event set. If a rule is determined not to have been triggered, record all its core information in detail, including the complete content of the triggered rule, the corresponding confidence level, the consequent defect type, and the time the rule was triggered (i.e., the current moment), ensuring the information is complete and traceable. If a rule is determined not to have been triggered, skip it and proceed to the next rule comparison. After all rule comparisons are completed, summarize and integrate all the records of triggered rules, arrange them according to the order of rule comparisons, and form the triggered rule list for the current window. Each record in this list corresponds to one triggered rule and includes core information such as the rule, confidence level, and consequent defect type.

[0042] Step 444: Based on the list of triggered rules, group them according to the consequent defect type, and calculate the confidence level of all triggered rules under each defect type to obtain the confidence level set corresponding to each defect type. Based on the confidence level set, use a weighted average to calculate the initial prediction probability of each type of formed defect occurring in the current window. Specifically, this includes: First, taking out the list of triggered rules for the current window obtained in step 443, sorting out the consequent defect type of each triggered rule in the list, clarifying the types of defect types corresponding to all triggered rules, and ensuring that no predicted defect type is missed; then, grouping all rules in the list of triggered rules according to the different consequent defect types, that is, grouping all triggered rules with the same consequent defect type into one group. Rules within the same group correspond to the same molding defect, while different groups correspond to different molding defects. Grouping is strictly based on the type of defect following the defect, ensuring no mixed or omitted groups occur. After grouping, for each defect type, the confidence scores of all triggered rules within that group are calculated. These confidence scores are then aggregated to form a confidence set for that defect type. This set includes the confidence scores of all triggered rules under that defect type, ensuring no triggered rule is omitted and no confidence scores from other defect type groups are mixed in. For example, if three rules in the triggered rule list have the same defect type (shrinkage mark) with corresponding confidence scores of 0.8, 0.75, and 0.85, then these three confidence scores are... The values ​​are summarized to form a confidence set {0.8, 0.75, 0.85} corresponding to the shrinkage defect type. If a defect type has no corresponding triggered rule, that is, the confidence set corresponding to the defect type is empty, its initial prediction probability is set to 0 when calculating the initial prediction probability of the defect type. Subsequently, based on the confidence set corresponding to each defect type, a weighted average is used to calculate the initial prediction probability of the defect type appearing in the current window. The specific calculation process of the weighted average is as follows: First, a corresponding weight is assigned to each confidence value in the confidence set. The weight assignment is based on the weighted support of the rule corresponding to that confidence level during the filtering in step 3. The higher the weighted support, the greater the corresponding weight. The specific weight assignment is as follows. The weighting rules are consistent with those used in step 3 to calculate the weighted support, ensuring the rationality and uniformity of the weight allocation. After weight allocation, the product of each confidence level and its corresponding weight is calculated, i.e., confidence level multiplied by weight, to obtain the weighted value corresponding to each confidence level. Then, the weighted values ​​corresponding to all confidence levels in the confidence set of this defect type are summed to obtain the total weighted value. At the same time, the weights corresponding to all confidence levels in the confidence set are summed to obtain the total weight. Finally, the total weighted value is divided by the total weight, and the result is the initial prediction probability of this defect type. The initial prediction probability ranges from 0 to 1; the larger the value, the greater the probability that the defect type will occur. For example, the confidence set corresponding to a certain defect type is {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 1 ...Given the confidence scores of 8, 0.75, and 0.85, with corresponding weights of 0.3, 0.2, and 0.5 respectively, first calculate the product of each confidence score and its corresponding weight: 0.8 multiplied by 0.3 equals 0.24, 0.75 multiplied by 0.2 equals 0.15, and 0.85 multiplied by 0.5 equals 0.425. Then, sum these three weighted values: 0.24 + 0.15 + 0.425 equals 0.815, obtaining the total weighted value. Simultaneously, sum the three weights: 0.3 + 0.2 + 0.5 equals 1.0, obtaining the total weight. Finally, divide the total weighted value (0.815) by the total weight (1.0) to obtain the initial predicted probability of this defect type, which is 0.815. Following this method, calculate the initial predicted probability for each defect type sequentially, ultimately obtaining the initial predicted probabilities for all formed defect types within the current window, thus completing the initial predicted probability calculation.

[0043] Step 445 involves extracting real-time monitoring values ​​of three key process parameters from the online process data stream, and simultaneously acquiring screw torque and screw speed data from the online process data stream. Specifically, this includes: First, determining the specific types of the three key process parameters: melt temperature, injection pressure, and holding time. These three parameters are core process parameters affecting the molding quality of automotive injection molds and causing molding defects. This process must be completely consistent with the defined three key process parameters to ensure parameter type uniformity and avoid arbitrarily changing or adding key process parameters. Then, from the online process data stream collected in real-time in step 440, the real-time monitoring values ​​of these three key process parameters are filtered and extracted. During the extraction process, the real-time and accuracy of the data are ensured. The extracted values ​​are the real-time monitoring values ​​at every moment within the current window, without omitting any parameter data at any moment. Simultaneously, the timestamp corresponding to each value is retained to ensure accurate correspondence between parameter values ​​and time. The specific extraction process is as follows: From the continuous process parameter data in the online process data stream, find the parameter data labeled melt temperature, injection pressure, and holding time, and extract each moment one by one according to the order of the timestamps. After the monitoring values ​​are extracted, the values ​​of the three key process parameters are organized to form a real-time value sequence for each key process parameter. Each sequence contains the real-time monitoring value of the parameter at each moment within the current window and the corresponding timestamp. At the same time, screw torque data and screw speed data are screened and extracted from the continuous process parameter data of the online process data stream. The extraction process is consistent with the extraction process of the three key process parameters to ensure the real-time performance, accuracy and completeness of the data. The extracted values ​​are the real-time monitoring values ​​of screw torque and screw speed at each moment within the current window. The timestamp corresponding to each value is retained to avoid time misalignment affecting subsequent calculations. After extraction, the screw torque data and screw speed data are organized to form real-time value sequences of screw torque and screw speed. Each sequence contains the real-time monitoring value of the corresponding parameter at each moment within the current window and the corresponding timestamp. This ensures that the screw torsion angle parameter can be calculated based on the data of these two sequences. During the extraction process, various types of parameter data are strictly distinguished to avoid confusion and omission, ensuring that the data of the three key process parameters, screw torque and screw speed are completely extracted.

[0044] Step 446: Based on the screw torque data and screw speed data, the screw torsion angle parameter is obtained by calculating the ratio of torque to speed or the phase difference. Specifically, this includes: First, taking out the real-time screw torque value sequence and the real-time screw speed value sequence obtained in step 445, and synchronizing these two sequences to ensure that the timestamps of the two sequences are completely consistent, that is, the screw torque value and screw speed value at the same moment correspond one-to-one, avoiding deviations in the calculation results due to time misalignment; During the synchronizing process, if it is found that there is only a screw torque value and no screw speed value at a certain moment, or only a screw speed value and no screw torque value, the average value of the corresponding parameter values ​​of the two adjacent moments before and after that moment is used to supplement it. After supplementation, ensure that the timestamps of the two sequences are completely synchronized and the data is complete. Then, using the ratio of torque to speed or the phase difference, calculate the synchronized screw torque and screw speed data to obtain the screw torsion angle parameter at each moment. The calculation process is performed one by one for each moment within the current window, without missing any moment, ensuring the continuity and completeness of the screw torsion angle parameter. Either of the two calculation methods can be selected, with fixed calculation rules to ensure the consistency and accuracy of the results. The specific calculation process is as follows: The first calculation method (ratio of torque to speed) takes the real-time screw torque and screw speed values ​​for each moment, and uses the screw torque at that moment... The first calculation method divides the torque value by the screw speed value at that moment, and the quotient is the screw torsion angle parameter at that moment. During the calculation, it is crucial to ensure that the units of the torque and speed values ​​are consistent to avoid deviations in the calculation results due to unit inconsistencies. If the units are inconsistent, the two values ​​should be converted to a unified unit before performing the division. For example, if the screw torque value at a certain moment is 100 N·m and the screw speed value is 50 r / min, and the units are consistent, dividing 100 by 50 yields the screw torsion angle parameter of 2 at that moment. The second calculation method (phase difference between torque and speed) involves taking the real-time screw torque and screw speed values ​​for each moment, recording the phase information corresponding to the two parameter values, and calculating the phase difference between them. The difference between each phase information is the screw torsion angle parameter at that moment. During the calculation process, the value range and calculation standard of the phase information are fixed to ensure that the phase difference calculation rules are consistent at each moment. For example, if the phase of the screw torque is 30° and the phase of the screw speed is 15° at a certain moment, subtracting 15° from 30° gives the screw torsion angle parameter of 15° at that moment. After the screw torsion angle parameter at each moment is calculated, the screw torsion angle parameters at all moments are arranged in chronological order according to the timestamps to form a real-time numerical sequence of screw torsion angle. This sequence contains the screw torsion angle parameter and corresponding timestamp at each moment in the current window, fully representing the real-time change of the screw torsion angle in the current window.

[0045] Step 447: Based on the real-time monitoring values ​​of three key process parameters and the screw torsion angle parameter, calculate the mean, variance, and range of each parameter within the sliding time window as process characteristic indicators. Based on these indicators, extract local fluctuation characteristics reflecting parameter fluctuations to obtain process stability characteristics. Specifically, this includes: First, setting the length of the sliding time window, which is shorter than the set time window length. Combining the fluctuation frequency of the screw torsion angle and the three key process parameters, as well as the response time of defect formation, ensures accurate extraction of local fluctuation characteristics of the parameters. The length of the sliding time window is fixed and does not change over time. Simultaneously, the sliding step size is set to one acquisition time, i.e., every interval of one acquisition time... At each step, the sliding time window is moved forward by one time point to ensure that the parameter fluctuation characteristics within each local time period can be continuously extracted. Then, the real-time numerical sequences of the three key process parameters (melt temperature, injection pressure, and holding time) extracted in step 445 and the real-time numerical sequence of the screw torsion angle calculated in step 446 are taken out, for a total of four parameter numerical sequences. For the numerical sequence of each parameter, the mean, variance, and range are calculated one by one within the sliding time window. These three values ​​together serve as the process characteristic index of the parameter. The four parameters are calculated separately without interference. The calculation process is performed one by one for each sliding time window to ensure that the process characteristic index within each sliding window can be accurately calculated.

[0046] The specific calculation process is as follows: For the numerical sequence of a specific parameter, within the current sliding time window, the parameter values ​​at all times within that window are extracted and summed together. Then, this sum is divided by the number of parameter values ​​within the sliding time window (i.e., the number of data collection times within that window). The quotient is the mean of the parameter within the current sliding time window. The mean reflects the average level of the parameter within that local time period. For example, if there are five melt temperature values ​​within a sliding time window: 190℃, 192℃, 191℃, 193℃, and 194℃, these five values ​​are first summed: 190 + 192 + 194℃. 1 plus 193 plus 194 equals 960℃. Dividing 960℃ by 5 gives the average melt temperature within the sliding window as 192℃. For variance calculation, given a sequence of values ​​for a specific parameter, within the current sliding time window, first calculate the mean value of that parameter within that window. Then, subtract the mean value from the parameter value at each moment within the window to obtain the deviation value for each value. Next, square each deviation value to obtain the square of each deviation value. Then, sum the squares of all deviation values ​​to obtain the sum of the squared deviation values. Finally, divide the sum of the squared deviation values ​​by the number of parameter values ​​within the sliding time window; the quotient is the value of that parameter within the current sliding time window. Variance reflects the degree of dispersion of a parameter's fluctuation within a local time period; the larger the variance, the more drastic the fluctuation. Range calculation, for a specific parameter's numerical sequence, involves extracting the parameter's values ​​at all times within the current sliding time window, identifying the maximum and minimum values, and then subtracting the minimum from the maximum. The difference is the range of the parameter within the current sliding time window, reflecting the maximum fluctuation range of the parameter within that local time period; the larger the range, the larger the fluctuation range. After calculating the mean, variance, and range of the four parameters (melt temperature, injection pressure, holding time, and screw torsion angle) across all sliding time windows, the range of each parameter is obtained. The corresponding process characteristic index sequence contains the mean, variance, and range of the parameter within each sliding time window. Subsequently, based on these process characteristic indices, local fluctuation features reflecting parameter fluctuations are extracted. These local fluctuation features include fluctuation frequency and fluctuation amplitude, which are completely consistent with the local fluctuation features. The extraction process of local fluctuation features is as follows: For the extraction of fluctuation frequency, for each parameter's process characteristic index sequence, the number of times the mean, variance, or range of the parameter changes significantly within a unit of time is counted. A significant change is defined as the amplitude of the change reaching a preset threshold. The number of such changes is the fluctuation frequency of the parameter, which reflects the frequency of parameter fluctuations.To extract the fluctuation amplitude, for each parameter's process characteristic index sequence, calculate the maximum range of that parameter across all sliding time windows, or calculate the maximum difference between the means of two adjacent sliding time windows. This value represents the fluctuation amplitude of the parameter, reflecting the severity of the parameter's fluctuation. The fluctuation frequencies and amplitudes of the four parameters are then summarized and integrated to form a process stability feature that comprehensively reflects the overall fluctuation of the process parameters within the current window. This process stability feature fully characterizes the local fluctuation state of the four core parameters.

[0047] Step 448: Obtain the pre-stored baseline features under normal production conditions, compare the process stability features with the baseline features, calculate the degree of deviation, and generate a dynamic correction coefficient based on the degree of deviation. Specifically, this includes: First, obtaining the pre-stored baseline features under normal production conditions. These baseline features are obtained under normal production conditions for automotive injection molds, without any molding defects. Following steps 445 to 447, multi-source process data under normal production conditions are collected, three key process parameters, screw torque, and screw speed data are extracted, the screw torsion angle parameter is calculated, process characteristic indicators are constructed, and local fluctuation features are extracted to form the process stability features. The baseline features are pre-stored in the data storage unit and include the fluctuation frequency and amplitude of the four core parameters (melt temperature, injection pressure, holding time, and screw torsion angle) under normal production conditions. The baseline features remain fixed. Subsequently, the process stability features obtained in step 447 within the current window... The process stability characteristics are compared one by one with pre-stored benchmark characteristics. The comparison process targets each local fluctuation characteristic (fluctuation frequency, fluctuation amplitude) of each core parameter, ensuring comprehensiveness and accuracy, and leaving no fluctuation characteristic overlooked. Specifically, for the fluctuation frequency of each core parameter, the deviation value is obtained by subtracting the fluctuation frequency of the parameter in the benchmark characteristics from the fluctuation frequency of the current process stability characteristics. For the fluctuation amplitude of each core parameter, the deviation value is obtained by subtracting the fluctuation amplitude of the parameter in the benchmark characteristics from the fluctuation amplitude of the current process stability characteristics. A positive deviation value indicates that the fluctuation frequency or amplitude of the current parameter is greater than the normal production state; a negative deviation value indicates that the fluctuation frequency or amplitude of the current parameter is less than the normal production state; a deviation value of 0 indicates that the fluctuation state of the current parameter is consistent with the normal production state.

[0048] After all deviation values ​​are calculated, the degree of deviation is calculated. The calculation process is as follows: First, a corresponding weight is assigned to each deviation value of each core parameter. The weight assignment is based on the influence of the parameter and its fluctuation characteristic on molding defects. The greater the influence, the greater the weight. The sum of the weights is 1, and the weights are preset and fixed. Then, the absolute value of each deviation value is multiplied by the corresponding weight to obtain the weighted deviation value for each deviation value (the absolute value is used to avoid positive and negative deviations canceling each other out, ensuring that the degree of deviation accurately reflects the overall deviation situation). Next, the weighted deviation values ​​corresponding to all fluctuation characteristics of all core parameters are added together to obtain the weighted deviation sum. This weighted deviation sum is the degree of deviation between the current process stability characteristic and the baseline characteristic. The greater the degree of deviation, the greater the deviation between the current process fluctuation state and the normal production state, and the greater the possibility of molding defects. The smaller the degree of deviation, the closer the current process fluctuation state is to the normal production state, and the lower the possibility of molding defects. Finally, based on the calculated degree of deviation, a dynamic correction coefficient is generated. The value range of the dynamic correction coefficient is between 0.8 and 1.2. The rules are preset and fixed, and the specific generation rules are as follows: When the deviation is 0, it means that the current process state is completely consistent with the normal production state, and the dynamic correction coefficient is set to 1.0, that is, no correction is made to the initial predicted probability; when the deviation is greater than 0 and less than the preset first deviation threshold, it means that the current process fluctuation is small and the deviation from the normal production state is not significant, and the dynamic correction coefficient is set between 1.0 and 1.1. The larger the deviation, the closer the correction coefficient is to 1.1; when the deviation is greater than or equal to the first deviation threshold and less than or equal to the preset second deviation threshold, it means that the current process fluctuation is large and the deviation from the normal production state is significant, and the dynamic correction coefficient is set between 1.1 and 1.2. The larger the deviation, the closer the correction coefficient is to 1.2; when the deviation is greater than the second deviation threshold, it means that the current process fluctuation is drastic and the deviation from the normal production state is extremely large, and the dynamic correction coefficient is directly set to 1.2; if the deviation is negative, that is, the current process fluctuation is less than the normal production state, it means that the current process state is more stable, and the dynamic correction coefficient is set between 0.8 and 1.0. The larger the absolute value of the deviation, the closer the correction coefficient is to 0.8. After the dynamic correction coefficient is generated, the coefficient value is retained.

[0049] Step 449: Obtain the initial prediction probability, add the dynamic correction coefficient to the initial prediction probability to obtain the corrected final prediction probability, and compare the final prediction probability with the preset warning threshold. When the final prediction probability exceeds the preset warning threshold, issue a warning signal corresponding to the predicted defect type. Specifically, this includes: First, taking out the initial prediction probabilities corresponding to all forming defect types within the current window calculated in step 444, clarifying the initial prediction probability value for each defect type to ensure that no defect type is missed and the initial prediction probabilities of different defect types are not confused. At the same time, taking out the dynamic correction coefficient generated in step 448. This dynamic correction coefficient is a unified value used to correct the initial prediction probability of all defect types to ensure the uniformity of the correction rules. Then, adding the dynamic correction coefficient to the initial prediction probability of each defect type to obtain the corrected final prediction probability. The addition method is multiplicative addition, that is, multiplying the initial prediction probability of each defect type by the dynamic correction coefficient, and the product is the defect type. In the calculation of the final predicted probability, it is ensured that the initial predicted probability and dynamic correction coefficient are accurate, the multiplication is correct, and the final predicted probability remains between 0 and 1. If the calculated result is greater than 1, the final predicted probability is set to 1; if the calculated result is less than 0, the final predicted probability is set to 0. For example, if the initial predicted probability of a certain defect type is 0.8 and the dynamic correction coefficient is 1.1, multiplying 0.8 by 1.1 yields a final predicted probability of 0.88 for that defect type; if the initial predicted probability of a certain defect type is 0.3 and the dynamic correction coefficient is 0.9, multiplying 0.3 by 0.9 yields a final predicted probability of 0.27 for that defect type. The final predicted probability of each defect type is calculated sequentially using the above method to ensure that the final predicted probability of all defect types is accurately corrected. Then, a preset warning threshold is obtained. The warning threshold is a fixed value set in advance, based on the quality standards of automotive injection molded products, historical defect occurrences, and production safety requirements. The value range of the warning threshold is between 0.5 and 0.A threshold of 8 is used to determine whether a warning signal needs to be issued. The warning threshold is fixed and does not change with the current predicted probability. The final predicted probability of each defect type is compared with the preset warning threshold one by one. The comparison process is performed separately for each defect type to avoid mixing comparison results from different defect types. The specific comparison and warning process is as follows: For a certain defect type, if the final predicted probability of that defect type is greater than the preset warning threshold, it indicates that the probability of that defect type occurring is high, exceeding the acceptable range, and a warning signal needs to be issued promptly to remind staff to take countermeasures. The specific form of the warning signal is preset, including one or more of the following: audible warning, visual warning, and text prompt warning. The warning signal clearly indicates the predicted defect type and its final predicted probability to ensure that staff can quickly understand the reason for the warning and the level of risk. For example, if the preset warning threshold is 0.7 and the final predicted probability of a certain defect type is 0.88, which is greater than 0.7, then a warning signal corresponding to that defect type is issued, indicating to staff that there is a high risk of that defect type and suggesting checking process parameters and equipment status.

[0050] If the final predicted probability of a certain defect type is less than or equal to the preset warning threshold, it indicates that the probability of that defect type occurring is low and within an acceptable range, so no warning signal is issued. After comparing the final predicted probabilities of all defect types with the warning threshold, the warning results are summarized. If the final predicted probabilities of multiple defect types exceed the warning threshold, a warning signal is issued for each defect type respectively, without omitting any high-risk defect type. If the final predicted probabilities of all defect types do not exceed the warning threshold, no warning signal is issued. At the same time, the final predicted probability, comparison results, and warning status for each instance are recorded to form a prediction and warning record.

[0051] It enables real-time prediction and timely early warning of molding defects, improves the accuracy and timeliness of defect prediction, effectively guides the quality control of automotive injection mold production, reduces molding defects, ensures product quality stability, and adapts to the actual production needs of automotive injection molding.

[0052] like Figure 2 As shown, embodiments of the present invention also provide a data processing system for correlation of molding defects in automotive injection molds, including: The data acquisition module is used to collect multi-source process data during the molding process of automotive injection molds, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set. The building module is used to set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and build the forming defect transaction dataset; The filtering module is used to filter frequent itemsets that meet the minimum support by statistically analyzing the frequency of each event itemset in the transaction dataset of molding defects. Based on the weighted support and confidence threshold, it extracts the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects, forming a defect association rule set. The matching module is used to perform rule matching on the online process data stream based on the defect association rule set, calculate the initial prediction probability of forming defects, extract the real-time monitoring values ​​of three key process parameters from the online process data stream, calculate the screw torsion angle parameter by solving the screw torque and speed data, construct process characteristic indicators and extract local fluctuation features to obtain process stability features, generate dynamic correction coefficients to correct the initial prediction probability, obtain the final prediction probability and generate corresponding early warning signals.

[0053] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0054] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for processing data related to molding defects in automotive injection molds, characterized in that, The method includes: Step 1: Collect multi-source process data of automotive injection mold during the molding process, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set; Step 2: Set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and construct the forming defect transaction dataset; Step 3: For the molding defect transaction dataset, the frequency of each event item set in the transaction is statistically analyzed to filter the frequent itemsets that meet the minimum support. Based on the weighted support and confidence threshold, the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects are extracted to form a defect association rule set. Step 4: Based on the defect association rule set, perform rule matching on the online process data stream to calculate the initial prediction probability of forming defects; extract the real-time monitoring values ​​of three key process parameters from the online process data stream, calculate the screw torsion angle parameter by solving the screw torque and speed data, construct process characteristic indicators and extract local fluctuation features to obtain process stability features, generate dynamic correction coefficients to correct the initial prediction probability, and obtain the final prediction probability to generate the corresponding early warning signal.

2. The method for processing correlation data of molding defects in automotive injection molds according to claim 1, characterized in that, Multi-source process data is collected during the molding process of automotive injection molds. Continuous process parameters are converted into discretized status event labels. Discrete alarm events and material changeover events are uniformly encoded into Boolean event items, generating a standardized event item set, including: Simultaneously acquire multi-source process data generated by automotive injection molds during at least one complete molding cycle to obtain the original multi-source dataset; Continuous process parameter data are extracted from the original multi-source dataset. The continuous parameter values ​​at each time step are converted into corresponding discrete state labels using a preset discretization rule to obtain the discretized representation of the continuous parameters. Each discrete state label is defined as an independent event item. Meanwhile, discrete equipment alarm event data and discrete material change record data are extracted from the original multi-source dataset. Each alarm type and each material change behavior is also defined as an independent event item, resulting in an initial event item set composed of all event items. Assign a unique event identifier to each independent event item in the initial event item set and encode it as a Boolean variable so that for any given time, if the event item occurs or exists, the Boolean value is set to 1, otherwise it is set to 0, resulting in a Boolean event item record with a timestamp. Based on timestamped Boolean event item records, the events are arranged in chronological order, and the generated event item sequence is cleaned to finally generate a standardized event item set.

3. The method for processing correlation data of molding defects in automotive injection molds according to claim 2, characterized in that, The multi-source process data includes at least continuous process parameter data obtained from real-time monitoring by sensors, discrete equipment alarm event data recorded by equipment controllers, and discrete material replacement record data recorded by the production management recording unit.

4. The method for processing correlation data of molding defects in automotive injection molds according to claim 3, characterized in that, Set the time window length, align and concatenate the standardized event itemset with the pre-recorded forming defect types according to the time window, and construct a forming defect transaction dataset, including: Based on the time window length, taking the occurrence time of each pre-recorded forming defect as the benchmark, all event items within the time window length are extracted from the standardized event item set to obtain the defect window event item set corresponding to each defect; Associate each defect window event item set with the corresponding formed defect type to form a defect window transaction with a defect type label; Based on the time window length, extract all event item sets within the time window that do not include the time when any forming defects occur from the standardized event item set, and uniformly mark these event item sets as defect-free type to obtain defect-free window transactions. All defective window transactions are merged with all non-defective window transactions to construct a complete defective transaction dataset. Each transaction includes a set of event items and a defect category label.

5. The method for processing correlation data of molding defects in automotive injection molds according to claim 4, characterized in that, For the molding defect transaction dataset, the frequency of each event itemset in the transaction is statistically analyzed to filter frequent itemsets that meet the minimum support. Based on weighted support and confidence thresholds, association rules between conditional events consisting of alarm event items and material change event items and molding defects are extracted to form a defect association rule set, including: The frequency of occurrence of each single event item in the statistical defect transaction dataset is counted, the support of each single event item is calculated, and all single event items with a support greater than or equal to the preset minimum support threshold are selected as frequent item sets. Based on frequent item sets, candidate item sets are generated by pairwise combinations. The support of each candidate item set in the formed defect transaction dataset is calculated, and item sets with support greater than or equal to the preset minimum support threshold are retained as frequent item sets. Based on frequent binary itemsets, candidate triple itemsets are generated through join operations. The support of each candidate triple itemset is calculated, and itemsets with support greater than or equal to a preset minimum support threshold are retained as frequent triple itemsets. This process is repeated until no new candidate itemsets can be generated, thus obtaining all frequent itemsets. For each frequent itemset, generate all possible association rules that may satisfy the conditions, where the condition event set consists of alarm event items and material change event items, and the defect type is the defect category label recorded in the transaction dataset; For each association rule, calculate the confidence score of each association rule, and assign corresponding weights to the event items involved in each rule based on the importance of the event, and calculate the weighted support score of each rule; The weighted support of each rule is compared with a preset weighted support threshold, and the confidence of each rule is compared with a preset confidence threshold. Only the association rules whose weighted support and confidence both exceed their respective thresholds are retained, and these are aggregated to form a defect association rule set.

6. The method for processing correlation data of molding defects in automotive injection molds according to claim 5, characterized in that, Based on the defect association rule set, rule matching is performed on the online process data stream to calculate the initial prediction probability of formed defects, including: Real-time acquisition of online process data streams; standardization of events in the data streams; conversion into standardized event items; formation of online event sequences. Based on the time window length, taking the current moment as the benchmark, all standardized event items falling within the time window are extracted from the online event sequence to form the current window event item set; Obtain the defect association rule set. Each association rule includes the antecedent condition event set, the consequent defect type, and the confidence level corresponding to the rule. The current window event item set is compared one by one with the antecedent condition event set of each rule in the defect association rule set to determine whether the current window event item set completely includes all event items in the antecedent condition event set of the rule; if it does, the rule is determined to be triggered, the triggered rule, the confidence level of the rule, and the consequent defect type of the rule are recorded to obtain the list of triggered rules for the current window. Based on the list of triggered rules, the rules are grouped according to the type of consequent defect. The confidence scores of all triggered rules under each defect type are calculated to obtain the confidence score set corresponding to each defect type. Based on the confidence score set, the initial prediction probability of each type of formed defect occurring in the current window is calculated using a weighted average.

7. The method for processing correlation data of molding defects in automotive injection molds according to claim 6, characterized in that, Real-time monitoring values ​​of three key process parameters are extracted from the online process data stream. The screw torsion angle parameter is calculated from the screw torque and speed data. Process characteristic indicators are constructed, and local fluctuation characteristics are extracted to obtain process stability features. Dynamic correction coefficients are generated to correct the initial prediction probability, resulting in the final prediction probability and corresponding early warning signals, including: Real-time monitoring values ​​of three key process parameters are extracted from the online process data stream, and screw torque and screw speed data are obtained from the online process data stream. Based on screw torque and screw speed data, the screw torsion angle parameters are obtained by calculating the ratio of torque to speed or the phase difference. Based on the real-time monitoring values ​​of three key process parameters and the screw torsion angle parameter, the mean, variance and range of each parameter are calculated within the sliding time window as process characteristic indicators. Based on the process characteristic indicators, local fluctuation characteristics reflecting parameter fluctuations are extracted to obtain process stability characteristics. Obtain the pre-stored baseline features under normal production conditions, compare the process stability features with the baseline features, calculate the degree of deviation, and generate dynamic correction coefficients based on the degree of deviation. The initial prediction probability is obtained, and the dynamic correction coefficient is superimposed on the initial prediction probability to obtain the corrected final prediction probability. The final prediction probability is then compared with the preset warning threshold. When the final prediction probability exceeds the preset warning threshold, a warning signal corresponding to the predicted defect type is issued.

8. The method for processing correlation data of molding defects in automotive injection molds according to claim 7, characterized in that, The three key process parameters include melt temperature, injection pressure, and holding time.

9. The method for processing correlation data of molding defects in automotive injection molds according to claim 8, characterized in that, The local fluctuation characteristics include fluctuation frequency and fluctuation amplitude.

10. A data processing system for correlation of molding defects in automotive injection molds, wherein the system implements the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect multi-source process data during the molding process of automotive injection molds, convert continuous process parameters into discrete status event labels, and uniformly encode discrete alarm events and material change events into Boolean event items to generate a standardized event item set. The building module is used to set the time window length, align and stitch the standardized event itemset with the pre-recorded forming defect types according to the time window, and build the forming defect transaction dataset; The filtering module is used to filter frequent itemsets that meet the minimum support by statistically analyzing the frequency of each event itemset in the transaction dataset of molding defects. Based on the weighted support and confidence threshold, it extracts the association rules between the conditional events consisting of alarm event items and material replacement event items and molding defects, forming a defect association rule set. The matching module is used to perform rule matching on the online process data stream based on the defect association rule set and calculate the initial prediction probability of the formed defect; Real-time monitoring values ​​of three key process parameters are extracted from the online process data stream. The screw torsion angle parameter is obtained by solving the screw torque and speed data. Process characteristic indicators are constructed and local fluctuation characteristics are extracted to obtain process stability characteristics. Dynamic correction coefficients are generated to correct the initial prediction probability and obtain the final prediction probability to generate corresponding early warning signals.

Images