Lithium battery diaphragm production quality tracing method and system based on internet of things

Abstract

The application discloses a lithium battery diaphragm production quality tracing method and system based on the Internet of Things, relates to the lithium battery diaphragm production quality tracing field, and the method comprises the following steps: collecting historical diaphragm production information and real-time diaphragm production information; determining key nodes of lithium battery diaphragm production, and setting a unique quality safety traceability code for each key node; combining the historical diaphragm production information and the quality safety traceability code to construct a quality anomaly knowledge graph; inputting the real-time diaphragm production information into the quality anomaly knowledge graph to obtain quality anomaly parameters; performing credibility verification on the quality anomaly parameters; determining the causes of the quality anomaly by using a root cause analysis model, and generating compensation parameters according to the causes of the quality anomaly; performing parameter compensation on the quality anomaly parameters based on the compensation parameters, and performing quality tracing in the quality anomaly knowledge graph according to the compensated quality anomaly parameters. The application can effectively improve the management efficiency of lithium battery diaphragm production.

Description

Technical Field

[0001] This application relates to the field of lithium battery separator production quality traceability, and in particular to a lithium battery separator production quality traceability method and system based on the Internet of Things. Background Technology

[0002] Lithium-ion batteries are the core energy carriers in new energy vehicles, energy storage devices, and consumer electronics. Their safety, stability, and consistency directly affect the performance of end products. The lithium-ion battery separator, as a key internal component, plays a crucial role in isolating the positive and negative electrodes and allowing ions to pass through. Its thickness uniformity, pore size distribution, and mechanical strength, among other parameters, have a decisive impact on the battery's cycle life, charge / discharge efficiency, and safety.

[0003] Currently, quality traceability in lithium-ion battery separator production relies primarily on manual records and localized information systems. From a production process perspective, lithium-ion battery separator production encompasses multiple key steps, including polymerization, extrusion, stretching, coating, slitting, and quality inspection. However, data from each step is stored independently. This independent storage creates "information silos," resulting in a lack of real-time interconnection between lithium-ion battery separator production processes. When conducting comprehensive analysis and traceability of lithium-ion battery separator production quality, it is difficult to obtain a complete data chain, making it impossible to accurately pinpoint the location of anomalies in the production process, thus impacting the management efficiency of lithium-ion battery separator production. For example, in actual production, temperature changes in the extruder may affect the stretching rate in the subsequent stretching process, which in turn affects the coating effect of the coating liquid on the separator. Simultaneously, different batches of materials may lead to variations in the composition of the coating liquid, thus affecting the quality of the separator. However, due to the lack of effective cross-process correlation analysis methods, when a quality anomaly occurs in the separator, it is difficult to quickly and accurately pinpoint which process or parameter is causing the problem, resulting in low efficiency in quality traceability and impacting the management efficiency of lithium-ion battery separator production.

[0004] There is currently no good solution to the above problems. Summary of the Invention

[0005] This application provides a method and system for quality traceability in lithium battery separator production based on the Internet of Things, which can improve the management efficiency of lithium battery separator production.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0007] Firstly, an IoT-based method for tracing the production quality of lithium battery separators is provided, the method comprising:

[0008] Collect historical and real-time diaphragm production information;

[0009] Based on the historical separator production information, key nodes in lithium battery separator production are identified, and a unique quality and safety traceability code is assigned to each key node.

[0010] By combining the historical diaphragm production information and the quality and safety traceability code, and using knife-cut Bayesian estimation and prior algorithms, a quality anomaly knowledge graph is constructed.

[0011] The real-time diaphragm production information is input into the quality anomaly knowledge graph to obtain quality anomaly parameters;

[0012] The reliability of the aforementioned quality anomaly parameters is verified;

[0013] If the quality anomaly parameters pass the credibility verification, the root cause analysis model is used to determine the cause of the quality anomaly, and compensation parameters are generated based on the cause of the quality anomaly.

[0014] The quality anomaly parameters are compensated based on the compensation parameters, and the quality anomaly is traced in the quality anomaly knowledge graph based on the compensated quality anomaly parameters.

[0015] In one possible implementation of the first aspect, the key nodes include production quality nodes and production process nodes. The step of determining the key nodes in lithium battery separator production based on the historical separator production information and assigning a unique quality and safety traceability code to each key node includes the following steps:

[0016] Select production parameter data and quality indicators from the historical diaphragm production information;

[0017] The correlation coefficient between the production parameter data and the quality index was calculated using the Pearson correlation coefficient.

[0018] The production node corresponding to the production parameter data whose correlation coefficient exceeds the preset correlation coefficient is taken as the production quality node.

[0019] The historical diaphragm production information is used to determine the diaphragm production process flow and the corresponding process steps and process nodes for each process flow.

[0020] A process flow diagram is drawn for each process flow, combining the process steps and process nodes corresponding to each process flow.

[0021] Determine the completion time for each of the aforementioned process nodes;

[0022] The production process nodes are determined based on the process flow chart and the completion time.

[0023] Assign a unique quality and safety traceability code to each of the key nodes.

[0024] In one possible implementation of the first aspect, the step of combining the historical diaphragm production information and the quality and safety traceability code and constructing a quality anomaly knowledge graph using cut-and-grab Bayes estimation and prior algorithms includes the following steps:

[0025] Extract a quality anomaly knowledge ontology from the historical diaphragm production information, and construct a quality anomaly knowledge base based on the quality anomaly knowledge ontology;

[0026] The pre-defined knowledge extraction model is used to extract the quality anomaly subject and quality anomaly object from the quality anomaly knowledge base.

[0027] The quality anomaly subject, the quality anomaly object, and the key node are stored in the form of triples;

[0028] The association relationships between all the triples are extracted using the knife-cut Bayes estimation and prior algorithm to generate association rules;

[0029] A quality anomaly knowledge graph is constructed by combining the association rules and the triples.

[0030] In one possible implementation of the first aspect, the step of extracting the association relationships between all the triples using knife-cut Bayesian estimation and prior algorithms to generate association rules includes the following steps:

[0031] The triples are transformed into transaction datasets, where each transaction dataset represents a quality anomaly case.

[0032] The transaction datasets are structured into itemset elements, wherein each item variable in the itemset elements corresponds to each transaction dataset;

[0033] For any given item variable, calculate the number of times the item variable appears in all elements of the itemset to obtain the empirical probability;

[0034] The bias and variance of the empirical probability are calculated using the knife-cut Bayesian estimation, and the empirical probability is adjusted using the bias and variance to obtain the probability estimate.

[0035] The probability estimates are combined with prior algorithms to select multiple frequent sets;

[0036] Filter out all non-empty proper subsets in the frequent sets, and generate candidate association rules between the frequent sets and the corresponding non-empty proper subsets;

[0037] The confidence level of each candidate association rule is calculated using a preset confidence level formula, and the candidate association rule whose confidence level is greater than or equal to a preset confidence level threshold is taken as the association rule.

[0038] In one possible implementation of the first aspect, the quality anomaly parameters include process anomaly parameters and equipment anomaly parameters, and the credibility verification of the quality anomaly parameters includes the following steps:

[0039] When the quality anomaly parameter is the process anomaly parameter, the process parameters corresponding to all the key nodes within the preset timestamp and the quality and safety traceability code are used as the process combination parameter set;

[0040] Within a preset time period, a sliding window hash value for the process combination parameter set is calculated using a hash function;

[0041] The sliding window hash value is compared with the baseline process hash value. If the sliding window hash value is not equal to the baseline process hash value, then it is determined that the quality anomaly parameter has passed the credibility verification.

[0042] In one possible implementation of the first aspect, the method further includes:

[0043] When the quality anomaly parameter is the equipment anomaly parameter, determine the initial state parameter set of all lithium battery separator production equipment, and generate an initial hash value based on the initial state parameter set;

[0044] The device status hash value of the device combination parameter set after each production task of the lithium battery separator production equipment is completed is calculated by a hash function, wherein the device combination parameter set is composed of the batch number of the production task and the current device status parameter set of the lithium battery separator production equipment.

[0045] The device status hash value is compared with the baseline device hash value. If the device status hash value is not equal to the baseline device hash value, then the quality anomaly parameter is determined to have passed the credibility verification.

[0046] In one possible implementation of the first aspect, the step of determining the cause of quality anomalies using a root cause analysis model and generating compensation parameters based on the cause of quality anomalies includes the following steps:

[0047] The quality anomaly parameters are input into the root cause analysis model to determine the causes of quality anomalies in the production of the lithium battery separator.

[0048] The causes of the quality anomalies are input into a preset compensation model to generate compensation parameters.

[0049] In one possible implementation of the first aspect, the step of inputting the quality anomaly parameters into the root cause analysis model to determine the causes of quality anomalies in the lithium battery separator production includes the following steps:

[0050] During the production process of the lithium battery separator, the quality abnormality phenomenon corresponding to the quality abnormality parameter will be regarded as the top event.

[0051] The top event is input into the fault tree logic chain of the root cause analysis model, and Boolean algebraization is used to identify the minimal cut set of the top event, wherein the minimal cut set consists of basic events;

[0052] The basic event probability of each basic event and the minimum cut set probability of each minimum cut set are calculated based on a preset probability formula.

[0053] The probability of the top event is calculated by combining the inclusion-exclusion principle with the minimum cut set and the probability of basic events.

[0054] The contribution of each basic event probability to the probability of the top event is calculated using a preset importance analysis algorithm.

[0055] The basic event corresponding to the largest contribution is taken as the cause of the quality anomaly.

[0056] Secondly, this application provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described IoT-based lithium battery separator production quality traceability method.

[0057] Thirdly, this application provides an Internet of Things-based lithium battery separator production quality traceability system, comprising:

[0058] The memory is configured to store instructions; and

[0059] The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the above-described IoT-based lithium battery separator production quality traceability method.

[0060] The above technical solution achieves several key benefits. First, by assigning unique quality and safety traceability codes to critical nodes in lithium-ion battery separator production and combining historical and real-time production information, precise traceability of the entire production process can be achieved. This facilitates rapid identification of the root cause of quality issues, improving problem-solving efficiency. Second, constructing a quality anomaly knowledge graph using knife-cut Bayesian estimation and prior algorithms systematically integrates and analyzes various data from the production process, enabling more accurate identification of potential quality anomalies and helping to prevent quality problems in advance, thus reducing quality risks. Third, inputting real-time separator production information into the quality anomaly knowledge graph allows for the real-time acquisition of quality anomaly parameters, facilitating timely detection of quality problems during production, preventing escalation, and minimizing losses. Validating the credibility of the quality anomaly parameters ensures the accuracy of monitoring results. If the quality anomaly parameters pass the credibility verification, a root cause analysis model is used to determine the causes of the quality anomalies and generate compensation parameters, which helps to deeply analyze the root causes of the problems and provides a basis for developing effective improvement measures. Compensation parameters are applied to quality anomaly parameters, and the resulting quality anomaly data is used for source tracing within a quality anomaly knowledge graph. This facilitates timely adjustments to the production process, optimizes manufacturing techniques, and improves product quality. Furthermore, source tracing clarifies responsibility and promotes standardized production management. Real-time monitoring and tracing of quality information throughout the production process enables timely identification and resolution of problems, preventing production delays and waste, and improving the management efficiency of lithium battery separator production.

[0061] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0062] Figure 1 A flowchart illustrating a method for quality traceability in lithium battery separator production based on the Internet of Things, provided for an embodiment of this application;

[0063] Figure 2 This is a schematic diagram of the structure of a quality anomaly knowledge graph provided in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0065] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0066] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0067] Figure 1 This illustration schematically depicts a process flow diagram of an IoT-based lithium battery separator production quality traceability method according to an embodiment of this application. Figure 1 As shown in the figure, this application provides a method for traceability of lithium battery separator production quality based on the Internet of Things, which may include the following steps.

[0068] S110. Collect historical and real-time diaphragm production information;

[0069] S120. Determine the key nodes in lithium battery separator production based on historical separator production information, and assign a unique quality and safety traceability code to each key node.

[0070] S130. Combine historical diaphragm production information and quality and safety traceability codes, and use knife-cut Bayes estimation and prior algorithms to construct a quality anomaly knowledge graph;

[0071] S140. Input the real-time diaphragm production information into the quality anomaly knowledge graph to obtain quality anomaly parameters;

[0072] S150. Verify the credibility of abnormal quality parameters;

[0073] S160. If the quality anomaly parameters pass the reliability verification, use the root cause analysis model to determine the cause of the quality anomaly, and generate compensation parameters based on the cause of the quality anomaly.

[0074] S170. Perform parameter compensation on the quality anomaly parameters based on the compensation parameters, and perform quality source tracing in the quality anomaly knowledge graph based on the compensated quality anomaly parameters.

[0075] First, historical diaphragm production information can be collected and organized using enterprise resource planning (ERP) systems or specialized data acquisition software. This historical information records various parameters and events from past production processes, helping to trace the source of quality problems. Real-time diaphragm production information can be collected using IoT technology. Sensors collect various parameters from the production process in real time, such as temperature, humidity, pressure, and speed. The collected data is transmitted via wired or wireless means to a data center or cloud platform for storage and processing. Real-time production information can promptly detect anomalies in the current production process, preventing problems from escalating.

[0076] Secondly, based on historical separator production information, key nodes in lithium-ion battery separator production are identified, and a unique quality and safety traceability code is assigned to each key node. In the lithium-ion battery separator production process, key nodes refer to those stages that have a significant impact on product quality, production efficiency, and cost control. These key nodes can be identified through in-depth analysis of historical separator production information. For example, the quality of raw materials directly affects the performance of the separator. Therefore, the procurement, inspection, and storage of raw materials are among the key nodes. During separator production, various raw materials need to be mixed and stirred to ensure their uniformity and consistency. The process parameters of this step have a significant impact on product quality, and the mixing and stirring step can also be considered a key node. After identifying each key node, to ensure the traceability of lithium-ion battery separator production quality, a unique quality and safety traceability code needs to be assigned to each key node. These traceability codes can be used to record production information, quality inspection results, and operators at each node, so that the source of the problem can be quickly located and appropriate measures taken when quality issues occur. In this embodiment, the quality and safety traceability code can be determined according to the actual situation and can be a QR code or a unique identifier for a production node in the production process. Based on the company's pre-set coding rules, a unique quality and safety traceability code is generated for each key node. The traceability code can include information such as the node name, production batch, and production date for better identification and management. During the production process, the traceability code is recorded on the production records and quality inspection reports at each node. Simultaneously, a traceability code database is established to enable rapid retrieval and traceability of relevant information when needed.

[0077] Subsequently, a quality anomaly knowledge graph is constructed by combining historical diaphragm production information and quality and safety traceability codes, and utilizing cut-and-grab Bayesian estimation and prior algorithms. In this embodiment, the cut-and-grab method is a resampling technique used to estimate the stability and bias of statistics. It can be used to estimate the degree of influence of different features on diaphragm quality anomalies. By repeatedly deleting a sample or a group of samples and recalculating the statistic, the cut-and-grab distribution of the statistic can be obtained, and its bias and variance can be estimated. Combined with Bayesian estimation, the cut-and-grab distribution can be used as prior information to update the posterior probability distribution. The prior algorithm is based on Bayes' theorem, using it to filter out multiple frequent sets, and calculating the confidence level of the association between each frequent set. Associations with confidence levels greater than a preset confidence threshold are used as associations between triples within the quality anomaly knowledge graph. A preset knowledge extraction model is used to extract quality anomaly subjects and objects from the quality anomaly knowledge base. Quality anomaly subjects can be the causes of quality anomalies, such as factors related to people, equipment, materials, processes, or the environment. An abnormal quality object refers to the specific object, product, or abnormal quality phenomenon affected by the abnormality. For example, in diaphragm production, the object might be the produced diaphragm product itself or its specific quality attributes. The pre-defined knowledge extraction model can be an attention-based knowledge extraction model. A quality abnormality knowledge graph is constructed by combining relationships, abnormal quality subjects, and abnormal quality objects.

[0078] After obtaining the quality anomaly knowledge graph, real-time diaphragm production information is input into it to obtain quality anomaly parameters. Real-time diaphragm production information refers to various data acquired in real-time during diaphragm production through various sensors, monitoring equipment, and manual recording. This includes real-time parameters of raw materials, the operating status of production equipment, production process parameters, and relevant operational information of operators. The quality anomaly knowledge graph is a knowledge network built based on historical data and domain knowledge, storing the relationships between various factors and quality anomalies in the diaphragm production process. Nodes in the graph can represent different production parameters, quality indicators, equipment status, raw material characteristics, etc., while edges represent the relationships between these entities, such as causal relationships and correlations. When real-time production information is input into the quality anomaly knowledge graph, the graph analyzes and infers from this real-time data based on its existing knowledge and relationship network. It compares real-time data with historical data to look for patterns or features similar to known quality anomalies. If certain production parameters are found to be outside the normal range, or if abnormal relationships exist between certain parameters, these parameters are identified as quality anomaly parameters. For example, if the knowledge graph records that a certain combination of raw material purity and equipment temperature is prone to causing uneven membrane thickness, and the real-time production information happens to show similar raw material purity and equipment temperature conditions, then these two parameters may be marked as quality anomalies, prompting operators to pay attention to the relevant links and take timely measures to avoid or correct quality problems.

[0079] Subsequently, the credibility of the quality anomaly parameters is verified. These parameters are various data indicators discovered during the production process that may be related to product quality issues. However, these parameters may contain errors or be misleading due to various reasons. Credibility verification first needs to determine whether these parameters are genuine and valid. In this embodiment, hash value verification is introduced to verify the credibility of the quality anomaly parameters. Introducing hash value verification into the credibility verification of quality anomaly parameters essentially uses cryptographic hashing technology to ensure the integrity and immutability of data during collection, transmission, and storage, thereby determining whether the parameters are genuine and valid. Specifically, a hash value is generated from the quality anomaly parameters collected in real time and compared with the hash values ​​of historical trusted data or blockchain-stored evidence. If they match, the data has not been tampered with and has high credibility; if they do not match, the data may have been contaminated or tampered with, requiring investigation.

[0080] After the quality anomaly parameters pass credibility verification, the root cause analysis model is used to determine the cause of the quality anomaly, and compensation parameters are generated based on the cause. In this embodiment, the root cause analysis model refers to a systematic approach to trace the root cause of the quality problem based on the quality anomaly parameters. First, by comparing multi-dimensional parameters of normal and abnormal production batches, the root cause analysis model is used to identify key factors strongly correlated with the quality problem and determine the root cause of the quality anomaly parameters. For example, the root cause analysis model reveals that the root cause of "abnormal rise in mold temperature" is "insufficient flow due to wear of the cooling water pump impeller," rather than a simple temperature setting error. After determining the cause of the quality anomaly, compensation parameters are generated based on the cause. The cause of the quality anomaly can be input into a preset parameter compensation model. In this embodiment, the preset parameter compensation model is a neural network model. The preset parameter compensation model determines which causes are the main factors leading to the quality anomaly. Based on the nature and degree of influence of the cause, corresponding compensation strategies are formulated and converted into specific compensation parameters. For example, if the root cause is unstable temperature control due to equipment aging, the compensation parameters may be adjusting the temperature controller setting or increasing the equipment maintenance frequency.

[0081] Subsequently, the quality anomaly parameters are compensated based on the compensation parameters, and the quality anomaly origin is traced in the quality anomaly knowledge graph based on the compensated quality anomaly parameters. Based on the compensation parameters generated from the previously analyzed causes of quality anomalies, the originally detected quality anomaly parameters are adjusted or corrected. For example, if the dimensional deviation of the product is found to be caused by excessive mold temperature, the compensation parameters generated based on root cause analysis might be to reduce the mold temperature setpoint and adjust the cooling water flow rate. The current abnormal mold temperature parameters and related cooling water flow rate parameters are adjusted based on the compensation parameters. The quality anomaly knowledge graph is a technology that organizes and stores various knowledge and information related to quality anomalies in the form of a graph. It includes various parameters in the production process, equipment information, process information, raw material information, and their interrelationships, as well as knowledge of past quality anomaly cases and their causes. Using the compensated quality anomaly parameters, related information and historical records are searched in the quality anomaly knowledge graph to trace the source and development path of quality problems. By querying and matching in the knowledge graph, it is possible to understand whether similar quality anomalies have occurred in the past, what the circumstances were, what factors caused them, and what measures were taken to resolve them. For example, after parameter compensation, quality anomaly parameters such as mold temperature may change. Inputting these compensated parameters into a quality anomaly knowledge graph for querying may reveal similar past cases with similar mold temperature and product dimensional deviations, ultimately tracing the cause back to a change in a certain characteristic of the raw materials. This allows for further analysis and control of this raw material characteristic, thereby achieving quality traceability and optimizing the production process.

[0082] By assigning unique quality and safety traceability codes to key nodes in lithium-ion battery separator production, and combining historical and real-time separator production information, precise traceability of the entire production process can be achieved. This helps to quickly locate the source of quality problems and improve problem-solving efficiency. Secondly, by constructing a quality anomaly knowledge graph using knife-cut Bayesian estimation and prior algorithms, various data from the production process can be systematically integrated and analyzed, thereby more accurately identifying potential quality anomalies and helping to prevent quality problems from occurring in advance, reducing quality risks. Subsequently, inputting real-time separator production information into the quality anomaly knowledge graph allows for the real-time acquisition of quality anomaly parameters, helping to promptly detect quality problems in the production process, preventing problems from escalating, and reducing losses. Validating the credibility of quality anomaly parameters ensures the accuracy of monitoring results. If the quality anomaly parameters pass the credibility verification, a root cause analysis model is used to determine the causes of quality anomalies and generate compensation parameters, which helps to deeply analyze the root causes of problems and provides a basis for developing effective improvement measures. Compensation parameters are applied to quality anomaly parameters, and the resulting quality anomaly data is used for source tracing within a quality anomaly knowledge graph. This facilitates timely adjustments to the production process, optimizes manufacturing techniques, and improves product quality. Furthermore, source tracing clarifies responsibility and promotes standardized production management. Real-time monitoring and tracing of quality information throughout the production process enables timely identification and resolution of problems, preventing production delays and waste, and improving the management efficiency of lithium battery separator production.

[0083] In one embodiment of this invention, the key nodes include production quality nodes and production process nodes. Determining the key nodes in lithium battery separator production based on historical separator production information and assigning a unique quality and safety traceability code to each key node includes the following steps:

[0084] S201. Select production parameter data and quality indicators from historical diaphragm production information;

[0085] S202. Calculate the correlation coefficient between production parameter data and quality indicators using the Pearson correlation coefficient.

[0086] S203. The production node corresponding to the production parameter data whose correlation coefficient exceeds the preset correlation coefficient shall be the production quality node.

[0087] S204. Determine the process flow of diaphragm production and the corresponding process steps and process nodes for each process flow by using historical diaphragm production information.

[0088] S205. Draw a process flow diagram for each process flow, combining the corresponding process steps and process nodes.

[0089] S206. Determine the completion time for each process node;

[0090] S207. Determine the production process nodes based on the process flow chart and completion time;

[0091] S208. Assign a unique quality and safety traceability code to each key node.

[0092] First, production parameter data and quality indicators from historical diaphragm production information are selected. In this embodiment, historical diaphragm production information refers to a large amount of records and data accumulated during the diaphragm production process, covering all aspects and stages of production, including production time, production equipment, raw material usage, and process operations. Production parameter data refers to various specific values ​​or variables used to describe and control the production process during diaphragm production. Examples include temperature, pressure, speed, raw material ratio, and additive dosage. These parameters directly affect the diaphragm production process and the performance of the final product. Quality indicators are a series of standards and parameters used to measure the quality of diaphragm products, such as thickness uniformity, porosity, air permeability, tensile strength, and puncture strength. By selecting these quality indicators, a link can be established between production parameters and product quality, allowing for research on how to improve product quality by optimizing production parameters, or identifying potential production parameter factors that may cause quality problems when they occur.

[0093] Next, the correlation coefficient between production parameter data and quality indicators is calculated using the Pearson correlation coefficient. The Pearson correlation coefficient is a statistical indicator used to measure the strength and direction of the linear relationship between two variables. It quantifies the closeness and direction of the linear relationship between these two variables. Specifically, after calculating the correlation coefficient, if it is close to 1, it indicates a strong positive linear correlation between the production parameters and quality indicators; that is, when production parameters increase, quality indicators tend to increase as well. If it is close to -1, it indicates a strong negative linear correlation; that is, when production parameters increase, quality indicators tend to decrease. If it is close to 0, it indicates a weak linear relationship, and there may be no significant linear correlation. Calculating this correlation coefficient helps analysts determine which production parameters have a significant impact on product quality indicators, and whether this impact is positive or negative, thus providing data support and decision-making basis for optimizing the production process and improving product quality. For example, if a certain production parameter is found to have a high Pearson correlation coefficient with a key quality indicator, then this parameter can be closely monitored and adjusted during the production process in order to achieve better quality control.

[0094] Production nodes corresponding to production parameter data with correlation coefficients exceeding a preset correlation coefficient are designated as production quality nodes. The correlation coefficient refers to the strength of the correlation between production parameters and quality indicators. In this embodiment, the preset correlation coefficient can be set according to the specific business needs of the enterprise. A production node refers to a key link or equipment in the production process, and each node corresponds to a set of production parameters. After calculating the correlation coefficient between production parameter data and quality indicators using the Pearson correlation coefficient, a specific value is set as the preset correlation coefficient. Then, the production links corresponding to production parameter data with calculated correlation coefficients greater than or equal to the preset correlation coefficient are identified as production quality nodes. In this embodiment, production quality nodes refer to those production links or process nodes that have a significant impact on the quality of the lithium battery separator during production. For example, if the preset correlation coefficient is set to 0.6, and among the calculated correlation coefficients between production parameters and quality indicators, the correlation coefficient of production parameter A is 0.7 and the correlation coefficient of production parameter B is 0.8, both exceeding the preset 0.6, then the production nodes corresponding to production parameters A and B, such as specific processing steps on the production line or a raw material addition step, will be identified as key nodes.

[0095] By analyzing historical diaphragm production information, the production process of diaphragms is determined, along with the corresponding process steps and process nodes for each process step. In this embodiment, the process flow refers to the complete production path from raw materials to finished products, consisting of a series of logically ordered process steps, specifically including dry processes, wet processes, and coating processes. A process step refers to a specific processing step within the process flow, serving as the basic unit constituting the flow (e.g., "stretching" is a step in a wet process). A process node refers to a key operational point or time point within a process step, typically corresponding to changes in material state, equipment operation, or quality control points (e.g., the time point when "the temperature rises to 150°C in the stretching process"). Based on historical diaphragm production information, past production processes are determined, clarifying the main stages from raw material preparation to finished diaphragm output. The sequence of each process step is arranged according to chronological order and logical relationships. For example, in a dry process, raw material preparation typically begins, followed by extrusion casting to form the base film, then longitudinal stretching, transverse stretching, and finally winding to obtain the finished diaphragm. Under each main process flow, specific process steps are further subdivided. Taking the wet process as an example, the "raw material preparation" step includes preparing polyolefin resin, organic solvents, etc.; the "solvent preparation" step involves mixing the organic solvent with additives; the "casting" step involves casting the prepared solution into sheets; the "longitudinal stretching" and "transverse stretching" steps involve stretching the cast sheets; the "extraction" step removes the solvent; and finally, there is the "winding" step. In each process step, key points that significantly affect diaphragm quality and production efficiency are identified as process nodes. In the "longitudinal stretching" step of the dry process, the settings of "stretching temperature," "stretching speed," and "stretching ratio" are key process nodes. The parameter control of these nodes directly affects the diaphragm's pore structure and mechanical properties, among other quality indicators.

[0096] By combining the process steps and process nodes corresponding to each process flow, a process flow diagram is drawn for each process flow. After obtaining the process steps and process nodes corresponding to each process flow based on historical diaphragm production information, professional flowchart drawing software can be used to draw the process flow diagram for each major process flow as a unit. Adjacent process rectangles are connected by arrows, and the direction of the arrows indicates the execution order of the processes. At the same time, the key node relationships between processes can also be represented by arrows or other connecting lines to clarify the input-output relationships between processes and the mutual influence of key nodes.

[0097] Determine the completion time of each process node. The completion time refers to the length of time from the start to the end of each process node, which is determined based on historical diaphragm production information. For example, in the diaphragm production process, the coating process starts from the moment the coating material is evenly applied to the base membrane, and ends when the coating operation is completed and the coated diaphragm reaches the specified degree of dryness or performance requirements. This time is the completion time of the coating process, which can be obtained through historical diaphragm production information.

[0098] Subsequently, production process nodes are determined based on the process flow chart and completion time. By analyzing the logical relationships and time constraints of each process node in the process flow chart, nodes with critical impact on the overall production process are identified. These nodes determine the efficiency and stability of the production process. Specifically, the steps include:

[0099] Determine the node relationships between process nodes in each process flow chart. Node relationships include parallel running nodes, sequential running nodes, and merging nodes.

[0100] Draw a diaphragm production flow chart based on the flow chart of each process and the relationship between nodes;

[0101] Based on the diaphragm production flow chart, node relationships, and completion time, calculate the earliest start time, earliest completion time, latest end time, and latest start time for each process node;

[0102] The process node whose earliest start time equals the latest start time and whose earliest finish time equals the latest finish time is designated as the production process node.

[0103] First, the node relationships between process nodes in each process flow diagram are determined. These relationships include parallel running nodes, sequential running nodes, and converging nodes. In this embodiment, parallel running nodes are process nodes that can be performed simultaneously on different equipment or production lines at the same time. Sequential running nodes refer to process nodes that must be performed in a specific order, where the output of one node is the input of the next, and the next node can only begin after the previous node has completed. Converging nodes refer to nodes where multiple parallel running nodes or multiple sequential running node paths converge to a common node, which must wait for all preceding nodes to complete before it can begin. In other words, it clarifies how these process nodes are interconnected and interact with each other during the production process. The node relationships mainly include parallel running nodes, sequential running nodes, and converging nodes.

[0104] Secondly, a diaphragm production flow chart is drawn based on the flowcharts of each process and the relationships between nodes. In this embodiment, the diaphragm production flow chart refers to integrating the individual process flowcharts into a complete diaphragm production flow chart. Specifically, firstly, the start and end points of the diaphragm production flow chart are determined based on the start and end points of each process flowchart; secondly, the order of each process flowchart is determined based on the start and end points of the diaphragm production flow chart; and the process flowcharts are arranged sequentially to obtain the diaphragm production flow chart. Parallel node relationships, sequential node relationships, and converging node relationships are then marked on the diaphragm production flow chart.

[0105] Based on the diaphragm production flow chart, node relationships, and completion times, the earliest start time, earliest finish time, latest finish time, and latest start time of each process node are calculated. First, the start and end nodes are determined in the diaphragm production flow chart, and the earliest start time of the start node is initialized. Starting from the start node, each process node is traversed, and the earliest start time and earliest finish time are calculated based on the node relationships of each process node. Specifically, when the node relationship is a parallel or sequential running node, the earliest start time of the process node is equal to the earliest finish time of its predecessor node. When the node relationship is a merging node, the maximum value of the earliest finish times of all predecessor nodes is used as the earliest start time. The earliest start time of each process node is calculated by adding its earliest start time to its corresponding finish time. Similarly, starting from the termination node, each process node is traversed, and the latest finish time and latest start time are calculated based on the node relationships. When the node relationships are parallel or sequential, the latest finish time of the process node is equal to the latest start time of its successor node. When the node relationships are merging nodes, the minimum of the latest start times of all successor nodes is taken as the latest finish time of the process node. The latest start time is obtained by subtracting the finish time from the latest finish time of each process node.

[0106] Starting from the starting node of the diaphragm production flow chart, the earliest start time (ES) of the starting node is set to 0. Next, the earliest finish time (EF) is calculated. The earliest finish time equals the earliest start time plus the duration of that node, i.e., EF = ES + duration. The earliest start time (ES) of subsequent nodes is calculated. Specifically, for sequentially running nodes, the earliest start time of a subsequent node equals the earliest finish time of its predecessor. For parallel running nodes, the earliest start time of all parallel nodes is the same, usually equal to the earliest finish time of the converging node. This calculation is repeated until the end point, calculating the earliest start time and earliest finish time of each node sequentially until the end of the flow chart. Then, the latest finish time and latest start time are calculated (in reverse traversal) to determine the latest finish time (LF) of the end node. The latest finish time of the end node of the flow chart is usually equal to its earliest finish time. The latest start time (LS) is calculated. The latest start time equals the latest finish time minus the duration of that node, i.e., LS = LF - duration. The latest finish time (LF) of the preceding node is calculated. For sequentially running nodes, the latest finish time of the preceding node equals the latest start time of the subsequent node. For nodes running in parallel, the latest finish time of all parallel nodes is the same, which is usually equal to the latest start time of the converging node. Repeat the calculation until the starting point, calculating the latest finish time and latest start time of each node in reverse order of the flowchart arrows, until the starting point of the flowchart.

[0107] Process nodes whose earliest start time equals their latest start time and whose earliest finish time equals their latest finish time are designated as production process nodes. The earliest start time refers to the earliest time a process can begin after all preceding processes are completed. The latest start time refers to the latest time a process must begin to ensure the entire project is completed on time. The earliest finish time refers to the earliest time a process can successfully complete after starting at its earliest start time, i.e., EF = ES + the process duration. The latest finish time refers to the latest time a process must complete to avoid impacting the overall project schedule. When the earliest start time equals the latest start time (ES = LS), it indicates that the start time of this process is critical; there is no buffer time, and it must begin strictly according to this time, otherwise it will affect the progress of the entire production process. When the earliest finish time equals the latest finish time (EF = LF), it indicates that the finish time of this process is critical; it must be completed within this time, with no room for delay. Process nodes that meet the above conditions are designated as production process nodes in the production process. Production process nodes are specific links or processes in the production process that have a critical impact on the progress and efficiency of the entire production process.

[0108] After identifying each critical node, assigning a unique quality and safety traceability code to each critical node refers to assigning a unique code to those key stages in the product manufacturing process that have a significant impact on product quality. This code is used to track and trace product quality and safety related information. The quality and safety traceability code contains specific information about that node, such as production time, production location, operators, and batches of raw materials used. By scanning or querying this traceability code, detailed production information of the product at that critical node can be obtained. Using the enterprise's production management information system or specialized traceability code generation software, a unique traceability code is generated for each critical node according to preset coding rules, and then associated and stored with the relevant information of that critical node, thus obtaining a unique quality and safety traceability code.

[0109] By assigning a unique quality and safety traceability code to each key node, detailed production information of the product at the key node can be quickly obtained through the quality and safety traceability code, determining whether the product meets quality standards, effectively improving the accuracy and efficiency of quality traceability, and helping to focus on monitoring and managing these key links.

[0110] In one embodiment of this invention, constructing a quality anomaly knowledge graph by combining historical diaphragm production information and quality and safety traceability codes, and utilizing knife-cut Bayesian estimation and prior algorithms, includes the following steps:

[0111] S310. Extract the quality anomaly knowledge ontology from the historical diaphragm production information, and construct a quality anomaly knowledge base based on the quality anomaly knowledge ontology.

[0112] S320. Extract the quality anomaly subject and quality anomaly object from the quality anomaly knowledge base using a preset knowledge extraction model.

[0113] S330. Store the quality anomaly subject, quality anomaly object, and critical node in the form of triples.

[0114] S340. Use the knife-cut Bayesian estimation and prior algorithm to extract the association relationships between all triples and generate association rules;

[0115] S350. Construct a quality anomaly knowledge graph by combining association rules and triples.

[0116] Figure 2 This application provides a schematic diagram of the structure of a quality anomaly knowledge graph, as shown in the embodiments below. Figure 2 As shown, multiple quality anomaly cases are constructed into a quality anomaly knowledge graph. These quality anomaly cases include extruder temperature anomalies, heat setting furnace pressure fluctuations, batch porosity exceeding standards, uneven lithium battery separator thickness, and lithium battery separator wrinkles. This figure is a structural schematic diagram and is for illustrative reference only.

[0117] A quality anomaly knowledge ontology is extracted from historical separator production information, and a quality anomaly knowledge base is constructed based on this ontology. In this embodiment, the quality anomaly knowledge ontology includes entity categories and relation categories. Entity categories represent all types, causes, and phenomena of quality defects during the lithium battery separator production process. Relationship categories represent the mapping between entity categories, such as the relationship between quality defects and quality defect phenomena, or the relationship between quality defects and quality defect parameters. From historical separator production information, the aforementioned entity and relation categories are identified through data mining, natural language processing, or manual analysis to obtain the quality anomaly knowledge ontology. After obtaining the quality anomaly knowledge ontology, it is stored according to a preset framework to form a searchable and analyzable database. For example, each anomaly record corresponds to an entity instance in the ontology, such as "On May 10, 2025, the temperature of coating machine No. 3 exceeded the standard (180℃ → standard 170℃), resulting in uneven separator thickness." Relationships between entities are linked through data association; for example, "temperature exceeded the standard" and "uneven thickness" are connected by the "caused" relationship.

[0118] Subsequently, a pre-defined knowledge extraction model is used to extract the quality anomaly subject and quality anomaly object from the quality anomaly knowledge base. In this embodiment, the quality anomaly subject refers to the entity that causes the quality anomaly. For example, a process-related quality anomaly subject could be excessive coating speed, abnormal slurry viscosity, or insufficient drying time; an equipment-related quality anomaly subject could be roller press bearing wear, sensor drift, or unstable scraper pressure. The quality anomaly object refers to the entity affected by the quality anomaly event. For example, a quality anomaly object could be uneven lithium battery separator thickness, excessive porosity, insufficient puncture strength, or surface scratches. The pre-defined knowledge extraction model can be natural language processing technology or a rule engine, combining the entity category and relation category of the quality anomaly knowledge ontology to extract the quality anomaly subject and quality anomaly object from the knowledge base. A pre-trained named entity recognition model can be used to identify the subject and object entities in the quality anomaly knowledge base. For example, anomaly records from the quality anomaly knowledge base can be input into a pre-trained named entity recognition model, such as "On May 14, 2025, wear of the scraper of the coating machine (equipment-related subject) caused coating peeling off the diaphragm of batch L20250514-001 (anomaly-related object)". The identified subject is "scraper wear" (equipment-related cause), and the identified object is "coating peeling off" (anomaly type).

[0119] After obtaining the quality anomaly subject and the quality anomaly object, the quality anomaly subject, the quality anomaly object, and key nodes are stored in the form of triples. A triple is the basic storage unit in a knowledge graph, consisting of a subject, a relation, and an object. The subject refers to the core object that causes the quality problem; the relation refers to the interaction between the subject and the object; and the object refers to the specific object affected by the quality anomaly. In this embodiment, the quality anomaly subject refers to the object or factor that causes the quality anomaly, such as equipment failure, raw material quality problems, improper process parameter settings, operator errors, etc., which are the active factors in the quality anomaly event. The quality anomaly object refers to the object affected or acted upon by the quality anomaly, which can be the quality characteristics of the product, a certain link in the production process, or the final product itself, such as dimensional deviations, substandard performance, appearance defects, or a problem in a certain process of the production flow. The relation represents the association or connection between the quality anomaly subject and the quality anomaly object, describing how the subject causes the object to have a quality anomaly. For example, words such as "cause," "lead to," "cause," "because," and "make" can all be used to express association relationships, reflecting the causal relationship or other logical connection between the subject and the object. By identifying key nodes, the root cause of a quality anomaly can be traced along its causal chain. For example, "insufficient raw material purity (the subject of the quality anomaly) leads to a decline in product performance (the object of the quality anomaly)." "Leads to" represents the correlation. Based on this relationship, when a decline in product performance is detected, the source factor—insufficient raw material purity—can be found by following the direction indicated by "leads to." The corresponding production process for this source factor can then be identified as a key node, thus pinpointing which production process is experiencing a problem. Triplet storage combines the subject of the quality anomaly, key nodes, and the object of the quality anomaly into a fixed structure for storing knowledge. Taking "equipment aging leads to product cracks" as an example, "equipment aging" is the subject of the quality anomaly, "product cracks" is the object of the quality anomaly, and "leads to" represents the correlation. By identifying the quality anomaly object, the quality anomaly subject, and the associated relationships, the equipment number or location of the equipment experiencing the anomaly can be located. The key nodes in the production process corresponding to the equipment experiencing the anomaly are then grouped into triplets. In this way, a large amount of complex quality anomaly information can be standardized and processed, making it easier for computers to store, query, and analyze, thus providing strong support for the diagnosis, prevention, and improvement of quality anomalies.

[0120] This study utilizes knife-cut Bayesian estimation and a prior algorithm to extract associations between all triples and generate association rules. Knife-cut Bayesian estimation combines the knife-cut method and Bayesian inference. The knife-cut method is a resampling technique that reduces estimation bias by sampling multiple subsets of the original data. Bayesian inference, based on Bayes' theorem, updates the estimates of unknown parameters using prior knowledge and observed data. In quality anomaly analysis, it is used to estimate the probability of associations between triples. The prior algorithm is based on the "support-confidence" framework and generates frequent itemsets through a layer-by-layer search. Specifically, knife-cut Bayesian estimation performs multiple samplings of the original quality anomaly data. After each sampling, Bayesian inference is used to calculate the probability of associations between triples. Then, based on the prior algorithm, association rules that meet certain conditions are extracted from these triples by calculating support and confidence. For example, by using Knife-Cut Bayes estimation, the probability of "raw material quality problems leading to product performance degradation" is relatively high. This is then combined with prior algorithms to calculate its support and confidence. If both meet preset standards, an association rule can be generated: "If there are raw material quality problems, then product performance degradation." These association rules can help companies quickly pinpoint the causes of quality anomalies and take corresponding measures to prevent and resolve quality issues, thereby improving product quality and production efficiency. Association rules are mined from data using Knife-Cut Bayes estimation and prior algorithms, revealing the relationships between triples. Through association rule mining, more complex association patterns between the quality anomaly subject, key nodes, and quality anomaly objects can be revealed, further enriching the content of the knowledge graph.

[0121] By combining association rules and triples to construct a quality anomaly knowledge graph, triples provide the most basic factual information for the knowledge graph. A large number of triples can comprehensively describe various phenomena and relationships in the quality anomaly domain, laying the foundation for knowledge graph construction. Association rules provide the knowledge graph with predictive and reasoning capabilities. Based on these rules, newly emerging quality anomalies can be predicted and analyzed, helping companies take preventative measures to avoid quality problems or quickly locate the cause and take corresponding solutions when quality anomalies occur. For example, if the knowledge graph contains the association rule "When the equipment operating temperature is too high, there is a high probability of product dimensional deviations," then when the equipment operating temperature is detected to be too high, the product dimensions can be checked in advance, or measures can be taken to reduce the equipment temperature to avoid quality anomalies. The extracted triples and association rules are used to construct the quality anomaly knowledge graph. Specialized knowledge graph construction tools or databases can be used to visualize the triples and association rules. For example, through a graphical interface, users can quickly find all quality anomaly subjects and key nodes related to a specific quality anomaly object, as well as the association paths and strengths between them.

[0122] By constructing a quality anomaly knowledge graph, we can intuitively display various entities and relationships related to quality anomalies, helping to identify quality anomaly issues in advance and effectively trace and resolve them. This enables proactive risk warnings for quality anomalies and effectively ensures the product quality of lithium battery separators.

[0123] In one embodiment of this example, the association relationships between all triples are extracted using knife-cut Bayesian estimation and prior algorithms to generate association rules, including the following steps:

[0124] S410. Transform the triples into transaction datasets, where each transaction dataset represents a quality anomaly case.

[0125] S420. Construct itemset elements from the transaction datasets, where each item variable in an itemset element corresponds to each transaction dataset.

[0126] S430. For any item variable, calculate the number of times the item variable appears in all itemset elements to obtain the empirical probability.

[0127] S440. Calculate the bias and variance of the empirical probability using the knife-cut Bayesian estimation, and adjust the empirical probability using the bias and variance to obtain the probability estimate.

[0128] S450. Combining probability estimates with prior algorithms to select multiple frequent sets;

[0129] S460. Filter out all non-empty proper subsets in the frequent sets and generate candidate association rules between the frequent sets and their corresponding non-empty proper subsets.

[0130] S470. Calculate the confidence level of each candidate association rule using a preset confidence level formula, and take the candidate association rule with a confidence level greater than or equal to the preset confidence level threshold as the association rule.

[0131] First, the triples are transformed into transaction datasets, where each transaction dataset represents a quality anomaly case. This transformation essentially converts the structured relational data in the knowledge graph into a standard format suitable for data mining algorithms, with the core purpose of adapting to data analysis needs and improving data utilization efficiency. A transaction dataset is a data organization format; in this embodiment, one transaction dataset represents a set of quality anomaly cases, such as extruder temperature anomalies, heat setting furnace pressure fluctuations, or excessive batch porosity. When transforming triples into transaction datasets, each quality anomaly case represented by a triple is expanded into a transaction containing more relevant information. For example, for the triple "excessive impurities in raw materials causing pores in the diaphragm," the transformed transaction dataset, in addition to the core element, may also include information such as the time of occurrence of the quality anomaly case, the production batch involved, the relevant production equipment, and the test results, forming a complete transaction.

[0132] Secondly, the transaction datasets are structured into itemset elements. Each item variable in an itemset element corresponds to a transaction dataset. A transaction dataset is a collection of multiple transactions, each of which can be viewed as a specific event or case. For example, in a quality anomaly scenario in lithium battery separator production, a transaction might represent a quality anomaly during a particular production process; multiple such transactions constitute a transaction dataset. The purpose of constructing itemset elements from the transaction datasets and assigning item variables to each transaction dataset is to transform specific quality anomaly cases into computable structured data to uncover potential association rules. An itemset element is a collection of all possible feature items extracted from the transaction dataset. These items can be various information related to the quality anomaly subject, key nodes, and quality anomaly objects. For example, from multiple quality anomaly transaction datasets, possible itemset elements include "equipment aging," "abnormal slurry viscosity," "trigger," and "product non-conforming." For each item in an itemset element, an item variable is generated. The item variable is a Boolean variable indicating whether the item appears in each transaction dataset. For example, if an itemset element is "device aging," in a given transaction dataset, if the transaction contains the information "device aging," then the corresponding item variable for "device aging" will have a value of 1; otherwise, it will have a value of 0. This method transforms the transaction dataset into a representation of itemset elements and item variables, making the data easier to use for association rule mining.

[0133] Next, for any given variable, calculate the number of times it appears in all itemset elements to obtain the empirical probability. In this embodiment, a variable refers to an independent element in the transaction dataset, such as an extruder temperature anomaly or a heat-setting furnace pressure fluctuation. An itemset element is a set of variables, such as {extruder temperature anomaly, heat-setting furnace pressure fluctuation}. The empirical probability is calculated by dividing the number of times the variable appears in all transactions by the total number of transactions. This frequency reflects the likelihood of the variable appearing in the dataset. The formula for calculating the frequency of a variable in all itemset elements is shown below:

[0134]

[0135] Iterate through all itemset elements. For each item variable, count the number of times it takes the value 1 across all transactions. For example, suppose there are 10 itemset elements, and the item variable corresponding to "equipment failure" takes the value 1 in 3 itemset elements. Then, "equipment failure" occurs 3 times. Divide the occurrence count of the item variable by the total number of transactions to obtain the empirical probability. Assuming the total number of transactions is 10, and "equipment failure" occurs 3 times, then the empirical probability of "equipment failure" is 3 ÷ 10 = 0.3, which is 30%. Empirical probability provides a quantitative indicator of the importance of item variables by calculating their frequency of occurrence across all itemset elements. Based on empirical probability, item variables can be ranked to identify those that occur frequently and are likely more important, helping to focus on these key items in subsequent analysis. In scenarios such as quality anomaly analysis and fault prediction, empirical probability can help identify item variables frequently associated with adverse events or faults, i.e., risk factors. Empirical probability can support decision-making. For example, in situations with limited resources, item variables with high empirical probabilities and high risks can be prioritized.

[0136] After obtaining the empirical probability, the bias and variance of the empirical probability are calculated using knife-cut Bayesian estimation. The empirical probability is then adjusted using the bias and variance to obtain the probability estimate. The knife-cut method is a resampling technique used to estimate the bias and variance of a statistic. It works by repeatedly removing one observation from the original data and calculating the statistic based on the remaining data, obtaining a series of statistic estimates. When calculating the bias and variance of the empirical probability, the knife-cut method can estimate the stability and accuracy of the empirical probability. For example, for the case where the calculated empirical probability is 0.3, the knife-cut method removes each item set element sequentially and recalculates the probability of that variable, obtaining a series of probability values. These probability values ​​can be used to calculate the bias and variance. The bias reflects the average difference between the empirical probability and the true probability, while the variance measures the dispersion of these probability estimates. Bayesian estimation is an estimation method that considers prior information. Based on past experience or knowledge, a prior estimate of the probability of a variable occurring is obtained, represented by a prior distribution. For example, based on the analysis of similar past data, it is believed that the probability of this variable occurring is more likely to be between 0.2 and 0.4, which can be described by a suitable prior distribution. Then, combined with the actually observed itemset data, Bayes' theorem is used to update the prior distribution to a posterior distribution, thus obtaining a more accurate probability estimate. This probability estimate integrates prior information and information from the observed data. Combining the knife-cut method with Bayesian estimation, the knife-cut method is first used to estimate the bias and variance of the empirical probability. Then, historical information is used to adjust some parameters in the Bayesian estimation, or the empirical probability is directly adjusted. For example, if the bias of the empirical probability estimated by the knife-cut method is large, it indicates that there may be some systematic error in the empirical probability. In the Bayesian estimation process, the prior or posterior distribution can be corrected according to the direction and magnitude of the bias, or the empirical probability can be directly adjusted to make it closer to the true probability value. The probability estimate obtained in this way considers the observed information of the data, utilizes prior knowledge, and corrects for the bias and variance of the empirical probability, making the empirical probability more accurate and reliable.

[0137] Subsequently, a prior algorithm is used to filter out frequent itemsets by combining probability estimates. The probability estimates, obtained through methods such as knife-cut Bayesian estimation, reflect the likelihood of a particular item or itemset appearing in the dataset. For example, in the quality anomaly data of lithium battery separator production, the probability estimate for the item "equipment temperature too high" is 0.3, indicating that the probability of this condition occurring among all quality anomalies is 0.3. These probability estimates provide fundamental information for subsequent frequent itemset filtering, helping to determine the likelihood of an itemset appearing. The prior algorithm is based on a "support-confidence" framework, generating frequent itemsets through a layer-by-layer search. Support represents the frequency of an itemset appearing in the dataset, while confidence represents the probability of another itemset appearing given the occurrence of one itemset. For example, for the itemset {equipment temperature too high, separator thickness uneven}, support is the proportion of this itemset appearing in all transaction datasets, and confidence can be the probability of uneven separator thickness occurring given a high equipment temperature. The prior algorithm starts with a single itemset, gradually generating multiple itemsets, and filters out frequent itemsets based on set support and confidence thresholds. Based on the set support and confidence thresholds, frequent binomial itemsets are selected. A binomial itemet is a set consisting of two items. If both the support and confidence of a binomial itemet exceed the threshold, it is retained as a frequent itemset; otherwise, it is discarded. For example, if the support threshold is set to 0.2 and the confidence threshold to 0.6, and the binomial itemet {equipment temperature too high, diaphragm thickness uneven} has a support of 0.3 and a confidence of 0.7, then it is considered a frequent itemset. In other words, a priori algorithm is used to determine whether the probability estimate is greater than or equal to the user-given minimum support threshold. If there are itemsets that are greater than or equal to the user-given minimum support threshold, then these itemsets are considered frequent itemsets.

[0138] After obtaining the frequent sets, all non-empty proper subsets within the frequent sets are filtered out, and candidate association rules between the frequent sets and their corresponding non-empty proper subsets are generated. In this embodiment, a frequent set refers to an itemset in a given transaction dataset whose frequency is greater than or equal to the minimum support threshold; a non-empty proper subset refers to a subset other than the empty set and itself. First, all frequent sets are traversed to find all non-empty proper subsets, which are subsets that do not contain the empty set (i.e., have at least one item) and are not equal to themselves. Next, for each non-empty proper subset, candidate association rules between the frequent sets and their corresponding non-empty proper subsets are generated. In this embodiment, the candidate association rules are expressions constructed based on the frequent sets and their non-empty proper subsets to describe the potential association relationships between data items. For example, if the frequent set F = {A, B, C} and its non-empty proper subset S = {A, B}, then the candidate association rule is {A, B} → {C}, meaning that if A and B appear, then C may also appear. These candidate association rules are summarized based on frequently occurring itemsets in the data. They represent potential associations and require further evaluation and screening using indicators such as support and confidence to determine their reliability and effectiveness in actual data.

[0139] The confidence level of each candidate association rule is calculated using a preset confidence level formula. Candidate association rules with a confidence level greater than or equal to a preset confidence level threshold are designated as association rules. In this embodiment, candidate association rules refer to the many possible association rules generated during the association rule mining process. These rules are called candidate association rules. A large number of possible rules are generated through association rule mining algorithms (such as Apriori and FP-Growth). Confidence level is an indicator of the reliability of an association rule, representing the proportion of a transaction containing the antecedent (A) that also contains the consequent (B). The formula for calculating the confidence level is as follows:

[0140]

[0141] Where Support(A∪B) is the support when A and B occur simultaneously, and Support(A) is the support when A occurs.

[0142] Confidence (A) B) represents the confidence level.

[0143] The confidence level of each candidate rule is calculated using a pre-set confidence level formula, and the calculated confidence level is compared with the pre-set confidence level threshold. If the calculated confidence level is greater than or equal to the pre-set confidence level threshold, these candidate association rules are adopted as association rules. In this embodiment, the pre-set confidence level threshold can be determined according to the actual situation.

[0144] By utilizing knife-cut Bayesian estimation and prior algorithms to extract the association relationships between all triples, the association patterns of abnormal events in production or services can be identified and root cause localization can be assisted. This allows for early warning of abnormal events in production and optimization of related warning processes, thereby improving the management efficiency of abnormal warnings.

[0145] In one embodiment of this invention, the quality anomaly parameters include process anomaly parameters and equipment anomaly parameters. The reliability verification of the quality anomaly parameters includes the following steps:

[0146] S510. When the quality abnormality parameter is a process abnormality parameter, the process parameters and quality and safety traceability codes corresponding to all key nodes within the preset timestamp are used as the process combination parameter set.

[0147] S520. Calculate the sliding window hash value of the process combination parameter set using a hash function within a preset time period;

[0148] S530. Compare the sliding window hash value with the baseline process hash value. If the sliding window hash value is not equal to the baseline process hash value, then the quality anomaly parameter is determined to have passed the credibility verification.

[0149] In this embodiment, quality anomaly parameters include process anomaly parameters and equipment anomaly parameters. Process anomaly parameters refer to abnormalities in parameters related to the production process, operating methods, and process control, which directly or indirectly lead to product quality problems. Equipment anomaly parameters refer to abnormalities in the operating status parameters of production equipment (such as coating machines, roller presses, slitting machines, etc.), leading to equipment malfunction or decreased accuracy, thereby causing quality problems. When the quality anomaly parameter is a process anomaly parameter, the process parameters and quality and safety traceability codes corresponding to all key nodes within a preset timestamp are used as a process combination parameter set. In this embodiment, the preset timestamp can be determined according to the actual situation. Process anomaly parameters refer to parameters that deviate from the normal range during the production process; that is, when the quality anomaly parameter is a process anomaly parameter, the process parameters and quality and safety traceability codes corresponding to all key nodes within the preset timestamp need to be integrated into a process combination parameter set. For example, the combination of "process code + parameter value + timestamp" can be used as the process combination parameter set.

[0150] Subsequently, within a preset time period, a sliding window hash value for the process combination parameter set is calculated using a hash function. Essentially, this uses hash technology to perform real-time fingerprinting of process parameters during production, used to monitor parameter fluctuations, trace data integrity, or detect abnormal changes. The preset time period is divided into multiple fixed-length sub-intervals, with the window moving at specific time intervals. For each sliding window's process combination parameter set, a unique value is calculated using a hash function, serving as the digital fingerprint of the window's data. This fingerprint is used to determine the hash value of the process combination parameter set within different time periods, thereby identifying any abnormal process parameters. In this embodiment, the sliding window hash value can be calculated every 5 seconds, covering parameter data from the most recent 10 minutes. If the window hash value suddenly changes, it indicates a significant fluctuation in the process parameters during that period, triggering an alert.

[0151] The sliding window hash value is compared with the baseline process hash value. If the sliding window hash value is not equal to the baseline process hash value, the credibility verification of the quality anomaly parameter is confirmed. In this embodiment, the baseline process hash value can be determined according to the actual situation. The obtained sliding window hash value is then compared with the baseline process hash value. When the sliding window hash value is not equal to the baseline process hash value, it indicates that the credibility of the current window process parameter having a quality anomaly parameter is high, confirming the existence of a process quality problem. Conversely, if the sliding window hash value is equal to the baseline process hash value, the credibility is low.

[0152] By using sliding window hash comparison, the precise time point of parameter mutation can be quickly identified, avoiding time-consuming manual investigation and effectively preventing data tampering from going undetected.

[0153] In one embodiment of this invention, the method further includes:

[0154] S610. When the quality anomaly parameter is an equipment anomaly parameter, determine the initial state parameter set of all lithium battery separator production equipment, and generate an initial hash value based on the initial state parameter set.

[0155] S620. Calculate the device status hash value of the device combination parameter set after each production task is completed by the lithium battery separator production equipment using a hash function. The device combination parameter set consists of the batch number of the production task and the current device status parameter set of the lithium battery separator production equipment.

[0156] S630. Compare the device status hash value with the baseline device hash value. If the device status hash value is not equal to the baseline device hash value, then the quality anomaly parameter is determined to have passed the credibility verification.

[0157] When the quality anomaly parameter is a device anomaly parameter, the initial state parameter set of all lithium battery separator production equipment is determined, and an initial hash value is generated based on the initial state parameter set. When a device anomaly parameter is detected, the credibility of the anomaly is verified by comparing the initial state hash value with the current state hash value, ensuring that the device state has not been tampered with or abnormally degraded. In this embodiment, the initial state parameter set refers to the set of baseline parameters of the device during normal operation, including hardware status, configuration parameters, environmental variables, etc. The initial hash value is generated by serializing the initial state parameter set and then calculating a fixed-length digest using a hash function, which serves as a "digital fingerprint" of the device state for subsequent comparison and verification. Parameters during normal device operation can be obtained from the device management system, the parameter set can be converted into a JSON string, and a hash function can be applied to generate the initial hash value.

[0158] The device status hash value of the equipment combination parameter set after each production task of the lithium battery separator production equipment is calculated using a hash function. The equipment combination parameter set consists of the batch number of the production task and the current equipment status parameter set of the lithium battery separator production equipment. Abnormal equipment parameters refer to quality anomaly signals caused by abnormal equipment operation during the lithium battery separator production process. The initial state parameter set refers to the set of key parameters of all lithium battery separator production equipment in its initial state. A unique initial hash value is obtained by calculating the initial state parameter set using a hash function. This value is equivalent to a "digital fingerprint" of the equipment status, used for subsequent comparison to check whether abnormal changes have occurred in the equipment's operating status. After each production task is completed, the equipment combination parameter set is collected, and the hash function is used again to calculate the equipment status hash value, which reflects the actual operating status of the equipment in a specific batch task.

[0159] Subsequently, the device status hash value is compared with the baseline device hash value. If the device status hash value is not equal to the baseline device hash value, the quality anomaly parameter is deemed to have passed the credibility verification. In this embodiment, the baseline device hash value can be determined according to the actual situation. The baseline device hash value is generated by hashing the actual status parameters of the device after completing a certain batch of production. The baseline device hash value refers to the standard fingerprint of the device's normal operation. The pre-set baseline hash value represents the parameter combination characteristics of the device in a normal state. Comparing the device status hash value with the baseline device hash value, if a certain batch hash value is not equal to the baseline device hash value, it indicates that the device status deviates from the normal range, and the credibility of determining that the device anomaly causes a quality problem is high.

[0160] By verifying the credibility of abnormal equipment parameters and using hash values ​​to sensitively reflect minute changes in equipment parameters, the production process can be ensured to strictly follow process standards. This allows for precise location and real-time early warning of equipment anomalies, ensuring the consistency of lithium battery separator production quality.

[0161] In one embodiment of this example, determining the causes of quality anomalies using a root cause analysis model and generating compensation parameters based on the causes of quality anomalies includes the following steps:

[0162] S710. Input the quality anomaly parameters into the root cause analysis model to determine the causes of quality anomalies in lithium battery separator production.

[0163] S720. Input the causes of quality anomalies into the preset compensation model to generate compensation parameters.

[0164] Quality anomaly parameters are input into the root cause analysis model to determine the causes of quality anomalies in lithium battery separator production. These parameters reflect specific data indicators of quality anomalies during lithium battery production. The root cause analysis model, built based on historical data, process knowledge, or algorithms, analyzes the causal relationship between "anomaly parameters" and "quality problems" to identify the root cause of the anomaly. Real-time or historical quality anomaly parameters are input into the model; for example, if excessive lithium battery separator thickness is detected, and large fluctuations in coating machine pressure are also observed, these two sets of data are used as input. The root cause analysis model then outputs the root cause of the quality anomaly.

[0165] Subsequently, the causes of the quality anomalies are input into a preset compensation model to generate compensation parameters. In this embodiment, the preset compensation model can be determined according to the actual situation. The preset compensation model is an algorithmic model built based on process knowledge or historical data, which can be a neural network model. It is used to generate adjustment parameters based on the causes, which are correction values ​​used to offset the impact of the anomalies. The preset compensation model calculates compensation parameters that can offset the impact of the causes of the anomalies according to preset rules or algorithms. For example, adjusting equipment parameters reduces the coating speed of the coating machine from 50 mm / s to 45 mm / s; correcting process parameters increases the drying temperature from 80℃ to 85℃, etc.

[0166] By leveraging the synergistic effect of root cause analysis models and compensation parameter generation mechanisms, accurate solutions can be developed to address quality issues in lithium-ion battery separator production, and accurate quality assurance solutions can be provided for lithium-ion battery separator production.

[0167] In one embodiment of this invention, inputting quality anomaly parameters into the root cause analysis model to determine the causes of quality anomalies in lithium battery separator production includes the following steps:

[0168] S810. During the production of lithium battery separators, the quality abnormality phenomenon corresponding to the quality abnormality parameter will be regarded as the top event.

[0169] S820. Input the top event into the fault tree logic chain of the root cause analysis model and use Boolean algebraization to identify the minimum cut set of the top event, where the minimum cut set consists of basic events.

[0170] S830. Calculate the basic event probability of each basic event and the minimum cut set probability of each minimum cut set based on the preset probability formula;

[0171] S840. Calculate the probability of the top event by combining the inclusion-exclusion principle with the minimum cut set and the probability of basic events.

[0172] S850: Calculate the contribution of the probability of each basic event to the probability of the top event by using a preset importance analysis algorithm;

[0173] S860. The basic event with the largest contribution is taken as the cause of the quality anomaly.

[0174] In the production of lithium-ion battery separators, the quality anomalies corresponding to abnormal parameters are designated as top events. These parameters are abnormal data detected during production; the anomalies are observable results such as separator wrinkles or insufficient permeability; and the top event is the core event at the top of the fault tree logic chain, requiring priority resolution. For example, insufficient separator permeability leading to a battery short circuit. The quality anomaly parameters are the quantitative manifestation of the top event. For instance, if the top event is unqualified separator thickness, the corresponding parameter is a measured thickness exceeding the standard range. Abnormal parameters are direct evidence of the top event, and monitoring these parameters can quickly trigger the fault tree analysis process. Designating quality anomalies as top events essentially uses a systematic fault analysis methodology to break down complex quality problems into traceable, quantifiable, and solvable underlying factors.

[0175] Subsequently, the top event is input into the fault tree logic chain of the root cause analysis model. Boolean algebraization is used to identify the minimal cut set of the top event, where the minimal cut set consists of basic events. The fault tree logic chain starts from the top-level fault of the system and decomposes it layer by layer into intermediate and bottom events through logic gates, reflecting the complete path of how the fault gradually propagates from basic events to system-level faults. It is a tool for analyzing fault propagation mechanisms and identifying key causes. The minimal cut set is the minimum combination of basic events in the fault tree that can lead to the occurrence of the top event. If any event in this combination does not occur, the top event will not occur. First, starting from the top event, possible direct causes are decomposed layer by layer, and then further decomposed into basic events to form a logic tree structure. Then, the logic gates in the fault tree are converted into Boolean expressions. For example, top event T = A ∧ B (AND gate, A and B occurring simultaneously leads to T); top event T = C ∨ D (OR gate, C or D occurring both lead to T). The expressions are simplified through Boolean algebra operations, and finally, a combination of several minimal cut sets is obtained. The minimal cut set is obtained through Boolean algebra simplification. For example, after simplification, we get T = (E∧F)∨G. Then the minimal cut sets are {E, F} and {G}, which represent that "E and F occurring simultaneously" will lead to the top event; "G occurring alone" will also lead to the top event. Each minimal cut set corresponds to a set of basic events that must exist simultaneously, which is a necessary and sufficient condition for the top event to occur.

[0176] Next, the basic event probability of each basic event and the minimum cut set probability of each minimum cut set are calculated based on preset probability formulas. The preset probability formulas are the first preset probability formula and the second preset probability formula. In this embodiment, the basic event probability refers to the probability of each bottom-level event (i.e., an event that cannot be further decomposed) occurring in the fault tree; the minimum cut set probability refers to the probability of each cut set occurring within all the minimum cut sets that lead to the top event; a minimum cut set is a set composed of a group of basic events. The basic event probability of each basic event can be calculated based on the first preset probability formula and historical data statistics. The first preset probability formula is:

[0177]

[0178] Where P(A) represents the probability of the basic event A occurring; n represents the number of times event A occurs during the observation period; and N represents the total number of trials or the total time period during the observation period.

[0179] The probability of a basic event is determined by calculating the probability of each basic event occurring within the historical data using a first preset probability formula.

[0180] The probability of each minimal cut set can be calculated using the second preset probability formula, which is:

[0181]

[0182] in It is the probability of the j-th minimum cut set. is the probability of the i-th basic event in the cut set, and r is the number of basic events in the cut set.

[0183] The minimum cut set probability of each minimum cut set is calculated using the second preset probability formula.

[0184] Utilizing the inclusion-exclusion principle combined with minimal cut sets and the probability of basic events to calculate the probability of the top event is a precise quantitative method for handling overlapping relationships between multiple minimal cut sets in fault tree analysis of lithium battery separator production. By progressively adding and subtracting the probabilities of overlapping events, duplicate calculations or omissions are avoided, thus accurately obtaining the probability of the union of all minimal cut sets, which is the total probability of the top event. The calculation process begins by calculating the probability of a single minimal cut set, followed by the probability of the intersection between minimal cut sets (two minimal cut sets occurring simultaneously, meaning their basic events occur together). Then, the inclusion-exclusion principle is applied layer by layer to calculate the union probability, and so on, until all possible intersection combinations are included, i.e., calculating the probability of the union of all minimal cut sets, yielding the total probability of the top event.

[0185] The contribution of each basic event probability to the probability of the top event is calculated using a pre-defined importance analysis algorithm. Importance analysis quantifies the contribution of each basic event to the probability of the top event, thereby identifying key components or weak points and providing a basis for optimizing system reliability. The pre-defined importance analysis algorithm can be probability importance (first-order importance), which calculates the first-order partial derivative of the basic event probability with respect to the top event probability. First-order importance directly reflects the marginal impact of changes in basic event probabilities on the probability of the top event. Subsequently, based on probability importance, the probability of the basic event itself is considered. Essentially, it is the product of "probability importance" and "basic event probability proportion," more realistically reflecting the actual risk contribution of basic events. Even if an event has a high probability importance, its actual contribution may be limited if its actual probability of occurrence is extremely low. For example, in lithium battery separator production, assuming the top event is "abnormal separator thickness," its fault tree includes the following basic event A: Insufficient raw material purity (probability p). A =0.1); B: Coating head temperature runaway (p B =0.05); C: Traction speed fluctuation (p C=0.08); through importance analysis, the probability importance is calculated as follows: IA=0.4, IB=0.5, IC=0.3; the critical importance is as follows: Icr,A=0.3, Icr,B=0.4, Icr,C=0.2. Therefore, the critical importance of coating head temperature runaway (B) is the highest, indicating that it contributes the most to the actual risk of abnormal diaphragm thickness, and the temperature control system should be optimized first.

[0186] The basic events with the highest contribution are identified as the causes of quality anomalies. Contribution refers to the degree of influence of a basic event on the probability of the top event occurring, calculated using an importance analysis algorithm. A higher value indicates a greater "contribution" of the basic event to the quality anomaly; that is, when the basic event occurs or changes, the probability of the top event occurring increases significantly. Quantitative analysis is used to identify the underlying factors that have the greatest impact on the probability of quality anomalies and these are recognized as the primary causes. Basic events refer to the most fundamental causes of quality anomalies (such as equipment failure, substandard raw materials, operational errors, etc.), and are events that cannot be further decomposed in the fault tree. The basic events with the greatest impact are selected as the causes of quality anomalies by ranking their contribution.

[0187] By identifying the causes of quality anomalies in lithium battery separator production, the most significant underlying events can be quickly pinpointed, avoiding misjudgments due to subjective experience. This allows for early warning of potential quality risks and enables preventative maintenance. Furthermore, the impact of each factor on quality anomalies can be quantified, providing data-driven support for resource allocation.

[0188] This application also provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described IoT-based lithium battery separator production quality traceability method.

[0189] This application also provides an IoT-based lithium battery separator production quality traceability system, including:

[0190] The memory is configured to store instructions; and

[0191] The processor is configured to retrieve instructions from memory and, when executing instructions, implement the aforementioned IoT-based lithium battery separator production quality traceability method.

[0192] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0193] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0194] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0195] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0196] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0197] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0198] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0199] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0200] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for quality traceability in lithium battery separator production based on the Internet of Things, characterized in that, The method includes the following steps: Collect historical and real-time diaphragm production information; Based on the historical separator production information, key nodes in lithium battery separator production are identified, and a unique quality and safety traceability code is assigned to each key node. By combining the historical diaphragm production information and the quality and safety traceability code, and using knife-cut Bayesian estimation and prior algorithms, a quality anomaly knowledge graph is constructed. The real-time diaphragm production information is input into the quality anomaly knowledge graph to obtain quality anomaly parameters; The reliability of the aforementioned quality anomaly parameters is verified; If the quality anomaly parameters pass the credibility verification, the root cause analysis model is used to determine the cause of the quality anomaly, and compensation parameters are generated based on the cause of the quality anomaly. Based on the compensation parameters, parameter compensation is performed on the quality anomaly parameters, and quality source tracing is performed in the quality anomaly knowledge graph based on the compensated quality anomaly parameters. The key nodes include production quality nodes and production process nodes. The step of determining the key nodes in lithium battery separator production based on the historical separator production information and assigning a unique quality and safety traceability code to each key node includes the following steps: Select production parameter data and quality indicators from the historical diaphragm production information; The correlation coefficient between the production parameter data and the quality index was calculated using the Pearson correlation coefficient. The production node corresponding to the production parameter data whose correlation coefficient exceeds the preset correlation coefficient is taken as the production quality node. The historical diaphragm production information is used to determine the diaphragm production process flow and the corresponding process steps and process nodes for each process flow. A process flow diagram is drawn for each process flow, combining the process steps and process nodes corresponding to each process flow. Determine the completion time for each of the aforementioned process nodes; The production process nodes are determined based on the process flow chart and the completion time. Assign a unique quality and safety traceability code to each of the key nodes; The quality anomaly parameters include process anomaly parameters and equipment anomaly parameters. The credibility verification of the quality anomaly parameters includes the following steps: When the quality anomaly parameter is the process anomaly parameter, the process parameters corresponding to all the key nodes within the preset timestamp and the quality and safety traceability code are used as the process combination parameter set; Within a preset time period, a sliding window hash value for the process combination parameter set is calculated using a hash function; The sliding window hash value is compared with the baseline process hash value. If the sliding window hash value is not equal to the baseline process hash value, it is determined that the quality anomaly parameter has passed the credibility verification. The method further includes: When the quality anomaly parameter is the equipment anomaly parameter, determine the initial state parameter set of all lithium battery separator production equipment, and generate an initial hash value based on the initial state parameter set; The device status hash value of the device combination parameter set after each production task of the lithium battery separator production equipment is completed is calculated by a hash function, wherein the device combination parameter set is composed of the batch number of the production task and the current device status parameter set of the lithium battery separator production equipment. The device status hash value is compared with the baseline device hash value. If the device status hash value is not equal to the baseline device hash value, then the quality anomaly parameter is determined to have passed the credibility verification.

2. The method according to claim 1, characterized in that, The step of combining the historical diaphragm production information and the quality and safety traceability code, and constructing a quality anomaly knowledge graph using cut Bayesian estimation and prior algorithms, includes the following steps: Extract a quality anomaly knowledge ontology from the historical diaphragm production information, and construct a quality anomaly knowledge base based on the quality anomaly knowledge ontology; The pre-defined knowledge extraction model is used to extract the quality anomaly subject and quality anomaly object from the quality anomaly knowledge base. The quality anomaly subject, the quality anomaly object, and the key node are stored in the form of triples; The association relationships between all the triples are extracted using the knife-cut Bayes estimation and prior algorithm to generate association rules; A quality anomaly knowledge graph is constructed by combining the association rules and the triples.

3. The method according to claim 2, characterized in that, The step of extracting the association relationships between all the triples using knife-cut Bayes estimation and prior algorithms to generate association rules includes the following steps: The triples are transformed into transaction datasets, where each transaction dataset represents a quality anomaly case. The transaction datasets are structured into itemset elements, wherein each item variable in the itemset elements corresponds to each transaction dataset; For any given item variable, calculate the number of times the item variable appears in all elements of the itemset to obtain the empirical probability; The bias and variance of the empirical probability are calculated using the knife-cut Bayesian estimation, and the empirical probability is adjusted using the bias and variance to obtain the probability estimate. The probability estimates are combined with prior algorithms to select multiple frequent sets; Filter out all non-empty proper subsets in the frequent sets, and generate candidate association rules between the frequent sets and the corresponding non-empty proper subsets; The confidence level of each candidate association rule is calculated using a preset confidence level formula, and the candidate association rule whose confidence level is greater than or equal to a preset confidence level threshold is taken as the association rule.

4. The method according to claim 1, characterized in that, The process of determining the causes of quality anomalies using a root cause analysis model and generating compensation parameters based on those causes includes the following steps: The quality anomaly parameters are input into the root cause analysis model to determine the causes of quality anomalies in the production of the lithium battery separator. The causes of the quality anomalies are input into a preset compensation model to generate compensation parameters.

5. The method according to claim 4, characterized in that, The step of inputting the quality anomaly parameters into the root cause analysis model to determine the causes of quality anomalies in the lithium battery separator production includes the following steps: During the production process of the lithium battery separator, the quality abnormality phenomenon corresponding to the quality abnormality parameter will be regarded as the top event. The top event is input into the fault tree logic chain of the root cause analysis model, and Boolean algebraization is used to identify the minimal cut set of the top event, wherein the minimal cut set consists of basic events; The basic event probability of each basic event and the minimum cut set probability of each minimum cut set are calculated based on a preset probability formula. The probability of the top event is calculated by combining the inclusion-exclusion principle with the minimum cut set and the probability of basic events. The contribution of each basic event probability to the probability of the top event is calculated using a preset importance analysis algorithm. The basic event corresponding to the largest contribution is taken as the cause of the quality anomaly.

6. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to execute the Internet of Things-based lithium battery separator production quality traceability method according to any one of claims 1 to 5.

7. A lithium battery separator production quality traceability system based on the Internet of Things, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the IoT-based lithium battery separator production quality traceability method according to any one of claims 1 to 5.

Images