A data intelligent management method and system for connector production

Abstract

The present application relates to the technical field of intelligent management, in particular to a data intelligent management method and system for connector production. The method comprises: constructing time sequence production transactions and calculating time weight for each production batch, pruning synchronously to obtain multi-constraint frequent item set when generating candidate set, generating multi-granularity candidate association rules and calculating confidence by cross-level association, calculating conditional probability of non-fulfillment of the latter under the condition of the former of each candidate association rule, nonlinear mapping of which as counterfactual stability score, constructing causal indicator, and weighting evaluation of candidate association rules by integrating confidence, causal indicator and counterfactual stability score of candidate association rules to screen out strong association rules. The scheme of the present application can quantify the time effectiveness of each production event on quality, distinguish real causal relationship from false statistical correlation, and reduce the risk of decision-making based on rules.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent management. More particularly, the present application relates to a data intelligent management method and system for connector production. BACKGROUND

[0002] As an indispensable component in electronic devices, connectors have characteristics of numerous procedures, complex process parameters, high precision requirements, etc. in the production and manufacturing process. In order to guarantee product quality, improve production efficiency and reduce cost, modern connector manufacturing industry is gradually transforming from relying on manual experience to data-driven intelligent management.

[0003] It has become a research hotspot in the industry to collect massive production process data in manufacturing execution system (MES), supervisory control and data acquisition (SCADA) and other systems, and analyze the correlation between process parameters and product quality by using data mining technology. Technical means include using association rule mining algorithms such as Apriori algorithm to discover potential relationships between production parameter combinations and quality results such as good products or defective products.

[0004] However, the strong correlation mined by the existing association rule may only be a statistical accompanying phenomenon, rather than a real causal relationship. For example, parameter A and quality defect B may be related to an unobserved hidden factor material batch C at the same time, resulting in a pseudo-association between parameter A and quality defect B. If the process parameters are adjusted based on this pseudo-association, it will not only fail to solve the problem, but may even have a negative impact. SUMMARY

[0005] The present application aims to provide a data intelligent management method and system for connector production to solve the technical problem that the statistical accompanying phenomenon under different rules is misjudged as a causal relationship in the prior art, resulting in increased decision-making risk. To this end, the present application provides solutions in the following two aspects.

[0006] In the first aspect, the present application provides a data intelligent management method for connector production, comprising the following steps:

[0007] The time series production transactions are constructed based on the quality detection results of the connector as the benchmark events, the time weight is calculated based on the time difference between each production parameter event and the benchmark event through a nonlinear decay function; a multi-dimensional constraint vector is set based on the weighted support, cost impact and process time fluctuation, and synchronous pruning is performed when generating the candidate set to obtain a multi-constraint frequent item set; the multi-constraint frequent item set is associated across levels according to a predefined production process ontology model to generate a multi-granularity candidate association rule and calculate the confidence of the candidate association rule; a conditional generation model based on the Transformer architecture is trained to learn the joint probability distribution of the production parameter events, and for each candidate association rule, the conditional probability that the posterior item is not true under the condition that the prior item is calculated by using the conditional generation model, and the nonlinear mapping of the conditional probability is taken as the counterfactual stability score; a causal indicator factor composed of rule time sequence consistency and potential mixed factor partial correlation is constructed, and the candidate association rules are weighted and evaluated by comprehensively considering the confidence of the candidate association rules, the causal indicator factor and the counterfactual stability score, and strong association rules representing key influencing factors of the production process are screened out.

[0008] Preferably, the time weight is calculated through a nonlinear decay function, including: taking the batch quality detection result as a benchmark event, and the occurrence time stamp of the event as ; for any production parameter event, the occurrence time stamp of the event is ; the time difference between the two is calculated ; an exponential decay function is used to calculate the time weight , wherein is a preset decay coefficient, is the time difference between the benchmark event and any production parameter event.

[0009] Preferably, the multi-dimensional constraint vector is set based on the weighted support, cost impact and process time fluctuation, and synchronous pruning is performed when generating the candidate set, including: setting a multi-dimensional constraint vector, the multi-dimensional constraint vector including a minimum weighted support threshold, a maximum cost impact threshold and a maximum process time fluctuation threshold; when generating a (k+1)-item set from a k-item set, for each newly generated candidate (k+1)-item set, the weighted support, cost impact value and process time fluctuation value are calculated in real time, and if the weighted support is lower than the minimum weighted support threshold, or the cost impact value is higher than the maximum cost impact threshold, or the process time fluctuation value is higher than the maximum process time fluctuation threshold, the candidate (k+1)-item set is directly pruned.

[0010] Preferably, the nonlinear mapping of the conditional probability as a counterfactual stability score comprises: for a candidate association rule A→B, calculating a conditional probability of the consequent B not being true given the antecedent A has occurred using the conditional generative model ; the counterfactual stability score is a monotonically decreasing function of the conditional probability .

[0011] Preferably, the constructing a causal indicator of the rule temporal consistency and the potential confounding factor partial correlation comprises: calculating a rule temporal consistency score : for a rule A→B, if the average timestamp of all events in the antecedent A is earlier than the timestamp of the consequent B event, then is 1, otherwise 0; identifying one or more potential confounding factors in the production process, and calculating a partial correlation coefficient between the antecedent A and the consequent B after controlling the potential confounding factors ; the causal indicator is obtained by weighted sum of the rule temporal consistency score and the absolute value of the partial correlation coefficient , and the formula is: wherein and are preset non-negative weight coefficients, is the rule temporal consistency score, is the partial correlation coefficient.

[0012] Preferably, the integrating the confidence of the candidate association rule, the causal indicator and the counterfactual stability score to perform a weighted evaluation on the candidate association rule, and screening out strong association rules representing key influencing factors of the production process comprises: for each candidate association rule, the calculation formula of the comprehensive evaluation score of the rule is: wherein Score is the comprehensive evaluation score of the rule, is the confidence of the candidate association rule, is the causal indicator, is the counterfactual stability score, , , are preset non-negative weight coefficients; when the Score of a rule is greater than a preset strong association rule threshold, the rule is determined as a strong association rule.

[0013] In a second aspect, a data intelligent management system for connector production comprises the following units:

[0014] The time weight calculation unit constructs a time sequence production transaction for each production batch of the connector based on a quality detection result of the connector as a reference event, and calculates a time weight based on a time difference between each production parameter event and the reference event through a nonlinear decay function; the generation unit of the association rule sets a multidimensional constraint vector based on weighted support, cost influence and process time fluctuation, synchronously prunes when generating a candidate item set, and obtains a multi-constraint frequent item set; the multi-granularity candidate association rule is generated by cross-level association of the multi-constraint frequent item set according to a predefined production process ontology model, and the confidence of the candidate association rule is calculated; the model generation unit trains a conditional generation model based on the Transformer architecture to learn the joint probability distribution of the production parameter events, calculates the conditional probability that the post-item of the rule is not true under the condition that the pre-item of the rule is true for each candidate association rule, and takes the nonlinear mapping of the conditional probability as the counterfactual stability score; the strong association rule screening unit constructs a causal indicator factor composed of rule time sequence consistency and potential mixed factor partial correlation, and screens out strong association rules representing key influencing factors of the production process by weighted evaluation of the confidence of the candidate association rule, the causal indicator factor and the counterfactual stability score.

[0015] Preferably, the time weight is calculated through a nonlinear decay function, including: taking the batch quality detection result as a reference event, and the occurrence time stamp of the event as ; for any production parameter event, the occurrence time stamp of the event is ; the time difference between the two is calculated ; an exponential decay function is used to calculate the time weight , wherein is a preset decay coefficient, is the time difference between the occurrence of the reference event and any production parameter event.

[0016] Preferably, the multidimensional constraint vector is set based on the weighted support, the cost influence and the process time fluctuation, and synchronous pruning is performed when generating a candidate item set, including: setting a multidimensional constraint vector, the multidimensional constraint vector including a minimum weighted support threshold, a maximum cost influence threshold and a maximum process time fluctuation threshold; when generating a (k+1)-item set from a k-item set, for each newly generated candidate (k+1)-item set, the weighted support, the cost influence value and the process time fluctuation value are calculated in real time, and if the weighted support is lower than the minimum weighted support threshold, or the cost influence value is higher than the maximum cost influence threshold, or the process time fluctuation value is higher than the maximum process time fluctuation threshold, the candidate (k+1)-item set is directly pruned.

[0017] Preferably, the nonlinear mapping of the conditional probability as a counterfactual stability score comprises: for a candidate association rule A→B, calculating the conditional probability that the consequent B does not hold given that the antecedent A occurs using the conditional generation model ; the counterfactual stability score is a monotonically decreasing function of the conditional probability .

[0018] The beneficial effects of the present application are: the present application can quantify the timeliness influence of each production event on quality by introducing a time weight based on a nonlinear decay function, thereby making up for the defect of ignoring the process timing; in the mining process, the present application integrates multi-dimensional business constraints such as cost influence and process time fluctuation, which ensures that the discovered rules not only have statistical significance, but also have actual economic and efficiency guiding value; by constructing a counterfactual stability score and a causal indicator, the present application conducts a deep causal inference test on the rules, which can distinguish real causal relationships from false statistical correlations, thereby reducing the risk of decision-making based on rules. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A step flowchart of a data intelligent management method for connector production in the embodiment is schematically shown;

[0020] Figure 2 A time weight decay function diagram in the embodiment is schematically shown. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0022] As shown in the drawings, Figure 1 a data intelligent management method for connector production in the embodiment comprises the following steps:

[0023] Step S1, for each production batch of the connector, constructing a time-ordered production transaction with the quality detection result of the connector as a reference event, and calculating a time weight through a nonlinear decay function based on the time difference between each production parameter event and the reference event.

[0024] The quality detection result of each production batch is taken as a reference event, such as qualified or unqualified. Each process parameter recorded in the production process of the batch is taken as a production parameter event, such as the pressure value of the stamping process and the temperature value of the injection molding process. The production parameter event and the reference event jointly constitute an event sequence sorted by time, for example, the transaction of batch B001 may contain {(stamping pressure = 150 MPa, t1), (plating current = 2.1 A, t2), (injection molding temperature = 285℃, t3), (final quality inspection = unqualified, )}.

[0025] The time weight is calculated by a nonlinear decay function, including: taking the batch quality detection result as a reference event, the occurrence time stamp of the event is ; for any production parameter event, the occurrence time stamp of the event is ; calculate the time difference between the two ; adopt an exponential decay function to calculate the time weight , wherein, is a preset decay coefficient, is the time difference between the occurrence of the reference event and any production parameter event.

[0026] The production parameter event closer to the time point of the occurrence of the quality problem has a greater possibility of affecting the quality problem, and therefore should be given a higher weight. The exponential decay function can well simulate the characteristics of rapid weakening of the influence over time, and compared with linear decay, it can more significantly distinguish events that are close in time and distant in time.

[0027] Suppose a batch of connectors is found to have quality defects at 10 o'clock in the morning, which is the reference event, and the influence weight of a plurality of production parameter events before this time needs to be analyzed. Event one is a flow anomaly that occurred at 9:55 am, and event two is a temperature fluctuation that occurred at 8 am. Assuming that the time difference unit is minutes, and the decay coefficient a is set to 0.1.

[0028] For event one, the time difference is 5 minutes, and the time weight is about 0.607; for event two, the time difference is 120 minutes, and the time weight is a value very close to zero, as shown in Figure 2 Through the above calculation, the focus of analysis can be automatically focused on the flow anomaly immediately before the defect occurs, and the temperature fluctuation two hours ago is basically ignored, making the correlation analysis more accurate and efficient.

[0029] Step S2, based on the weighted support, cost impact and process time fluctuation, a multi-dimensional constraint vector is set, synchronous pruning is performed when generating the candidate set, and a multi-constraint frequent item set is obtained; according to a pre-defined production process ontology model, the multi-constraint frequent item set is cross-level correlated to generate a multi-granularity candidate association rule and calculate the confidence of the candidate association rule.

[0030] The multi-dimensional constraint vector is set based on the weighted support, cost impact and process time fluctuation, and synchronous pruning is performed when generating the candidate set, including: setting a multi-dimensional constraint vector, which includes a minimum weighted support threshold, a maximum cost impact threshold and a maximum process time fluctuation threshold; when generating a (k+1)-item set from a k-item set, for each newly generated candidate (k+1)-item set, the weighted support, cost impact value and process time fluctuation value are calculated in real time, and if the weighted support is lower than the minimum weighted support threshold, or the cost impact value is higher than the maximum cost impact threshold, or the process time fluctuation value is higher than the maximum process time fluctuation threshold, the candidate (k+1)-item set is directly pruned.

[0031] In the early stage of association rule mining, parameter combinations that do not meet the actual business requirements are eliminated, avoiding wasting computing resources on a large number of useless candidate items. Not only the frequency of pattern occurrence (i.e. weighted support) is considered, but also the engineering and business considerations (i.e. the cost increase and the disturbance to the production rhythm that may be brought by adjusting the parameter combination) are integrated.

[0032] For example, in a production line, the minimum weighted support is set to 0.6, the maximum cost impact is 5 yuan, and the maximum process time fluctuation is 2 seconds. Assuming that there is a 2-item set that meets the conditions, i.e. humidity greater than 70% and the speed of mechanical arm A is 1.2 meters per second. Now try to add a third parameter, the static voltage is higher than 80 kilovolts, to generate a candidate 3-item set.

[0033] The indicators of the new 3-item set are calculated. The calculation shows that its weighted support is 0.7, and the cost impact is an increase of 4 yuan, both of which meet the requirements; but the process time fluctuation is calculated to be an increase of 2.5 seconds. Since 2.5 seconds exceeds the set maximum process time fluctuation threshold of 2 seconds, this candidate 3-item set will be immediately pruned even if the support and cost are qualified, and will not be used to generate more complex 4-item sets, thereby ensuring that the rules mined have practical application feasibility.

[0034] Step S3, training a conditional generation model based on the Transformer architecture to learn the joint probability distribution of production parameter events, for each candidate association rule, using the conditional generation model to calculate the conditional probability that the post-item is not true given the pre-item, and taking the nonlinear mapping of the conditional probability as the counterfactual stability score.

[0035] A decoder-only Transformer model similar to GPT is trained with event sequences in a time series as input. For a candidate rule, e.g. X as antecedent and Y as consequent, the event sequence corresponding to X is input into the trained model, and the conditional probability of the model generating a non-Y event (i.e. ¬Y) is calculated. The counterfactual stability score is derived from the conditional probability by a sigmoid function, the lower the probability value, the more stable the rule, the higher the score.

[0036] wherein the Transformer model is a conditional generation model, and its core structure is composed of stacked encoder and / or decoder blocks, each block mainly contains two sub-layers: a multi-head self-attention module and a position-wise feed-forward network. The multi-head self-attention mechanism calculates multiple sets of independent attention weights in parallel, so that the model can simultaneously pay attention to all other elements in the sequence when processing a certain element in the sequence, and dynamically assign different weights according to the relevance, thereby obtaining long-distance dependencies; while the feed-forward network further transforms and computes information on the output of the attention layer. In addition, residual connections and layer normalization are also incorporated to prevent gradient vanishing and accelerate model convergence, and position encoding is used to inject the time or position information of the element into the model. Preferably, the Transformer architecture includes a model with 6 layers of encoder and 6 layers of decoder, the vector dimension of each layer is 512, the attention mechanism has 8 heads, each head processes a 64-dimensional vector, and the hidden layer dimension of the feed-forward network is 2048.

[0037] The nonlinear mapping of the conditional probability is used as the counterfactual stability score, including: for a candidate association rule A→B, the conditional probability of the consequent B not being true under the condition that the antecedent A occurs is calculated using the conditional generation model ; the counterfactual stability score is a monotonically decreasing function of the conditional probability , used to represent the stability degree of the consequent also occurring when the antecedent occurs.

[0038] The confidence degree shows the proportion of B occurring when A occurs, but the counterfactual stability score focuses on how rare the exceptional cases of B not occurring when A occurs. The fewer the exceptional cases, the more stable the association between A and B, and the closer to a certain causal relationship.

[0039] Suppose there is a rule A→B in the production process, where A is the temperature exceeding 30 degrees Celsius and B is the yield reaching 99.5%. By analyzing historical data, it is found that there are 1000 batches of temperature exceeding 80 degrees Celsius. Among the 1000 times, 900 times reach 99.5% of the yield, and 100 times do not reach it.

[0040] At this time, the conditional probability of the former A occurring and the latter B not being true is 0.1. A simple monotonically decreasing function can be set to calculate the counterfactual stability score, for example, 1 minus the conditional probability is equal to 0.9, and the score of 0.9 indicates the stability of the rule. Compared with another rule with a stability score of 0.7, the former is obviously a more reliable process guide.

[0041] Step S4, constructing a causal indication factor composed of rule temporal consistency and potential confounding factor partial correlation, and comprehensively evaluating the candidate association rules by weighting the confidence of the candidate association rules, the causal indication factor and the counterfactual stability score, and screening out strong association rules representing key influencing factors of the production process.

[0042] Constructing a causal indication factor composed of rule temporal consistency and potential confounding factor partial correlation, comprising: calculating a rule temporal consistency score : for rule A→B, if the average time stamp of all events in the former A is earlier than the time stamp of the event in the latter B, then is 1, otherwise 0; identifying one or more potential confounding factors in the production process, and calculating the partial correlation coefficient between the former A and the latter B after controlling the potential confounding factors; the causal indication factor is obtained by weighted sum of the rule temporal consistency score and the absolute value of the partial correlation coefficient , and the formula is: , wherein and are preset non-negative weight coefficients, is the rule temporal consistency score, is the partial correlation coefficient.

[0043] The causal indication factor is to distinguish the more likely existing causal relationship from the statistical association, which includes two core evaluation criteria. First, the cause must occur before the result, which is the basic time rule of causal relationship, which is embodied by the rule temporal consistency score; second, the association of the two variables cannot be a pseudo-association driven by a third external factor, i.e. confounding factor, which is tested by the partial correlation coefficient. Among them, the partial correlation is to eliminate the common influence of variable Z (confounding factor) on variables A and B when measuring the linear relationship between variables A and B. In the present application, the confounding factors include but are not limited to temperature, humidity, etc.

[0044] The confidence of the candidate association rule, the causal indicator factor and the counterfactual stability score are combined to evaluate the candidate association rule, and strong association rules representing key influencing factors of the production process are screened out, including: for each candidate association rule, the calculation formula of the comprehensive evaluation score of the rule is: , wherein Score is the comprehensive evaluation score of the calculation rule, is the confidence of the candidate association rule, is the causal indicator factor, is the counterfactual stability score, , , is a preset non-negative weight coefficient; when the Score of a rule is greater than a preset strong association rule threshold, the rule is determined as a strong association rule.

[0045] The quality of the candidate rule is comprehensively evaluated, and multiple key attributes of the rule, including the universality of the rule, that is, the confidence, the causal credibility, that is, the causal indicator factor, and the certainty of the result, that is, the counterfactual stability score, are fused into a total score. By adjusting the weight coefficient, different types of rules can be focused on according to the needs.

[0046] For example, for a rule about adjusting the parameter A to reduce the part burr B, the indicators of the rule are calculated. The confidence of the rule is 0.85, which means that 85% of the time the parameter A is adjusted, the burr is reduced; the causal indicator factor of the rule is 0.8, which indicates that the causal chain of the rule is relatively reliable; and the counterfactual stability score of the rule is 0.95, which means that there are few exceptions of adjusting the parameter but not reducing the burr.

[0047] Suppose the company values the three equally, and sets the weight , , all as one-third. Then the comprehensive evaluation score is about 0.867. If the company sets the strong association rule threshold as 0.8, since 0.867 is greater than 0.8, this rule about the laser cutting parameter will be screened out as a high-value strong association rule and recommended to the process engineer.

[0048] The application also provides a data intelligent management system for connector production. It includes the following units:

[0049] A time weight calculation unit constructs a time sequence production transaction based on the quality detection results of the connector for each production batch of the connector, calculates the time weight through a nonlinear decay function based on the time difference between each production parameter event and the reference event.​​

[0050] The generation unit of the association rule sets a multi-dimensional constraint vector based on the weighted support, cost impact and process time fluctuation, performs synchronous pruning when generating the candidate item set, and obtains a multi-constraint frequent item set; according to a pre-defined production process ontology model, the multi-constraint frequent item set is cross-layer associated to generate a multi-granularity candidate association rule and calculate the confidence of the candidate association rule.

[0051] The model generation unit trains a conditional generation model based on a Transformer architecture to learn the joint probability distribution of the production parameter events, and for each candidate association rule, calculates the conditional probability that the post-item is not true under the condition that the pre-item is true by using the conditional generation model, and takes the nonlinear mapping of the conditional probability as the counterfactual stability score.

[0052] The strong association rule screening unit constructs a causal indicator factor composed of rule temporal consistency and potential confounding factor partial correlation, and performs weighted evaluation on the candidate association rule by comprehensively considering the confidence of the candidate association rule, the causal indicator factor and the counterfactual stability score, and screens out strong association rules representing key influencing factors of the production process.

[0053] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, including a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing the data intelligent management method for connector production according to the above-mentioned application.

[0054] The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface, the settings and functions of which are known in the art, and thus will not be described here.

[0055] In this description, the term "application" also means any computer program product storing such a program for use with or in connection with a computer system, apparatus or device. The program can be stored on any apparatus-readable medium, for example, but not limited to, any volatile memory or non-volatile memory. In this description, the term "memory" also means any computer program product storing such a program for use with or in connection with a computer system, apparatus or device. The program can be stored on any apparatus-readable medium, for example, but not limited to, any volatile memory or non-volatile memory. In this description, the term "computer-readable medium" means any tangible medium that stores, communicates, or otherwise provides data that can be used by an instruction execution system, apparatus or device. The computer-readable medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as, for example, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), and the like, or any other medium that can be used to store the desired information and that can be accessed by an application, module, or both. Any such computer storage media can be part of the device or accessible or connectable thereto. Any application or module described in this description can be implemented by computer-readable / executable instructions stored or otherwise held by such computer-readable media.

[0056] In the description of the present description, the meaning of "a plurality of" is at least two, for example, two, three or more, and the like, unless otherwise explicitly specified.

[0057] Although the present description has shown and described several embodiments of the present application, it will be apparent to those skilled in the art that many modifications, changes and substitutions can be made thereto without departing from the spirit and scope of the present application.

Claims

1. A data intelligent management method for connector production, characterized in that, Includes the following steps: For each production batch of connectors, a time-series production transaction is constructed with the connector quality inspection result as the benchmark event. The time weight is calculated using a nonlinear decay function based on the time difference between each production parameter event and the benchmark event. Multi-dimensional constraint vectors are set based on weighted support, cost impact, and process time fluctuations. Synchronous pruning is performed when generating candidate itemsets to obtain multi-constraint frequent itemsets. According to a predefined production process ontology model, cross-level association is performed on the multi-constraint frequent itemsets to generate multi-granularity candidate association rules and calculate the confidence of the candidate association rules. A conditional generation model based on the Transformer architecture is trained to learn the joint probability distribution of production parameter events. For each candidate association rule, the conditional generation model is used to calculate the conditional probability that the antecedent of the rule is not true under the condition that the consequent is true. The nonlinear mapping of the conditional probability is used as the counterfactual stability score. A causal indicator factor consisting of rule temporal consistency and partial correlation of potential confounding factors is constructed. The candidate association rules are weighted and evaluated by combining the confidence of the candidate association rules, the causal indicator factor and the counterfactual stability score, and strong association rules that characterize the key influencing factors of the production process are screened out.

2. The data intelligent management method for connector production according to claim 1, characterized in that, The calculation of time weights using a nonlinear decay function includes: The batch quality inspection results are recorded as the baseline event, and the timestamp of the event is... ; For any production parameter event, the event occurrence timestamp is: ; Calculate the time difference between the two ; Using an exponential decay function Calculate time weights ,in, The preset attenuation coefficient, This is the time difference between the baseline event and any production parameter event.

3. The data intelligent management method for connector production according to claim 1, characterized in that, The method of setting a multi-dimensional constraint vector based on weighted support, cost impact, and process time fluctuation, and performing synchronous pruning during candidate set generation, includes: Define a multidimensional constraint vector, which includes a minimum weighted support threshold, a maximum cost impact threshold, and a maximum process time fluctuation threshold; When generating (k+1)-itemsets from k-itemsets, for each newly generated candidate (k+1)-itemsets, the weighted support, cost impact value, and process time fluctuation value are calculated in real time. If the weighted support is lower than the minimum weighted support threshold, or the cost impact value is higher than the maximum cost impact threshold, or the process time fluctuation value is higher than the maximum process time fluctuation threshold, then the candidate (k+1)-itemsets is directly pruned.

4. The data intelligent management method for connector production according to claim 1, characterized in that, The step of using the nonlinear mapping of the conditional probability as a counterfactual stability score includes: For a candidate association rule A→B, the conditional generation model is used to calculate the conditional probability that the subsequent rule B does not hold given that the preceding rule A has occurred. ; The counterfactual stability score The conditional probability It is a monotonically decreasing function.

5. The data intelligent management method for connector production according to claim 1, characterized in that, The construction of the causal indicator factor, which consists of rule-based temporal consistency and partial correlation of potential confounding factors, includes: Calculate the timing consistency score of the rules For rule A→B, if the average timestamp of all events in the preceding term A is earlier than the timestamp of the subsequent term B, then It is 1 if it is true, otherwise it is 0. Identify one or more potential confounding factors in the production process, and after controlling for these potential confounding factors, calculate the partial correlation coefficient between the preceding term A and the following term B. ; The causal indicator By scoring the temporal consistency of the rules with partial correlation coefficient We obtain the weighted sum of the absolute values, as shown in the formula: ,in and These are the preset non-negative weighting coefficients. The score is based on the consistency of the rule sequence. This is the partial correlation coefficient.

6. The data intelligent management method for connector production according to claim 1, characterized in that, The candidate association rules are evaluated by weighting their confidence level, causal indicator factor, and counterfactual stability score to select strong association rules that characterize key influencing factors in the production process, including: For each candidate association rule, the formula for calculating the rule's comprehensive evaluation score is as follows: Where Score is the overall evaluation score calculated according to the rules. The confidence level of the candidate association rule. As a causal indicator, For counterfactual stability score, , , These are preset non-negative weighting coefficients; When a rule's score is greater than the preset threshold for strongly correlated rules, the rule is determined to be a strongly correlated rule.

7. A data intelligent management system for connector production, characterized in that, Includes the following units: The time weight calculation unit constructs a time-series production transaction for each production batch of connectors, using the connector quality inspection results as the benchmark event. Based on the time difference between each production parameter event and the benchmark event, it calculates the time weight through a nonlinear decay function. The association rule generation unit sets a multi-dimensional constraint vector based on weighted support, cost impact, and process time fluctuation. It performs synchronous pruning when generating candidate itemsets to obtain multi-constraint frequent itemsets. Based on a predefined production process ontology model, it performs cross-level association on the multi-constraint frequent itemsets to generate multi-granularity candidate association rules and calculates the confidence of the candidate association rules. The model generation unit trains a conditional generation model based on the Transformer architecture to learn the joint probability distribution of production parameter events. For each candidate association rule, the conditional generation model is used to calculate the conditional probability that the antecedent of the rule is false under the condition that the consequent is false. The nonlinear mapping of the conditional probability is used as the counterfactual stability score. The strong association rule screening unit constructs a causal indicator factor composed of rule temporal consistency and partial correlation of potential confounding factors. It comprehensively evaluates the candidate association rules by weighting the confidence of the candidate association rules, the causal indicator factor, and the counterfactual stability score, and screens out the strong association rules that characterize the key influencing factors of the production process.

8. The data intelligent management system for connector production according to claim 7, characterized in that, The calculation of time weights using a nonlinear decay function includes: The batch quality inspection results are recorded as the baseline event, and the timestamp of the event is... ; For any production parameter event, the event occurrence timestamp is: ; Calculate the time difference between the two ; Using an exponential decay function Calculate time weights ,in, The preset attenuation coefficient, This is the time difference between the baseline event and any production parameter event.

9. The data intelligent management system for connector production according to claim 7, characterized in that, The method of setting a multi-dimensional constraint vector based on weighted support, cost impact, and process time fluctuation, and performing synchronous pruning during candidate set generation, includes: Define a multidimensional constraint vector, which includes a minimum weighted support threshold, a maximum cost impact threshold, and a maximum process time fluctuation threshold; When generating (k+1)-itemsets from k-itemsets, for each newly generated candidate (k+1)-itemsets, the weighted support, cost impact value, and process time fluctuation value are calculated in real time. If the weighted support is lower than the minimum weighted support threshold, or the cost impact value is higher than the maximum cost impact threshold, or the process time fluctuation value is higher than the maximum process time fluctuation threshold, then the candidate (k+1)-itemsets is directly pruned.

10. The data intelligent management system for connector production according to claim 7, characterized in that, The step of using the nonlinear mapping of the conditional probability as a counterfactual stability score includes: For a candidate association rule A→B, the conditional generation model is used to calculate the conditional probability that the subsequent rule B does not hold given that the preceding rule A has occurred. ; The counterfactual stability score The conditional probability It is a monotonically decreasing function.

Images