Liquid crystal display screen anti-static foam box production line monitoring method

Abstract

The invention discloses a liquid crystal display screen anti-static foam box production line monitoring method, and relates to the technical field of production monitoring. The method comprises the following steps: collecting historical data of each workshop section, preprocessing the historical data, and constructing a historical data set; extracting the dynamic time sequence characteristics, and generating a mutual inductance rule base for revealing the causal relationship between the parameters based on the dynamic time sequence characteristics; the production line is monitored in real time, and the root section and parameters of the process abnormality are positioned by matching the real-time features with the rule base; performing simulation intervention on a shadow production line in the anti-static foam box production line according to the root section and the parameters, if the simulation intervention effect reaches an expected value, performing intervention adjustment on a main production line of the anti-static foam box production line, and performing collaborative optimization on the safety of the main production line through a hierarchical model prediction control technology; and performing static safety load verification on the product to form closed-loop quality control. According to the invention, the conversion from passive deviation rectification to beforehand prevention is realized, and the product consistency and the production efficiency are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of production line monitoring technology, specifically to a method for monitoring a production line of anti-static foam boxes for liquid crystal displays. Background Technology

[0002] LCD screens are extremely fragile during manufacturing, transportation, and assembly, and are easily damaged by electrostatic discharge (ESD), leading to bright spots, black spots, striped interference, and even driver chip malfunction. To ensure their safety, packaging materials with ESD protection are crucial. Anti-static foam boxes, with their conductive fillers, static dissipation layers, and cushioning structures, are key components for the packaging and handling of LCD screens. The protective performance of these foam boxes highly depends on the precise control and stable coordination of process parameters at each stage of production. Even minor deviations in any stage can cause the ESD protection function to fail to meet standards, thus posing a potential risk to the valuable display screen.

[0003] Currently, monitoring of such production lines typically relies on a traditional method based on sampling inspection of key indicators and adjustments made by human experience. This method primarily involves sampling and measuring key electrostatic safety indicators such as surface resistivity and charge decay time of finished foam boxes using equipment like electrostatic potential detectors after critical stages such as mixing and molding, or when products are off the production line. When anomalies are detected in the test data, engineers rely on their experience to manually analyze and adjust parameters in potentially related upstream stages to correct deviations. However, this traditional method, based on sampling inspection at the end of the production line, obtains static and delayed quality results, lacking the ability to uncover dynamic correlations and causal chains between parameters across multiple stages of the entire process. This results in corrective action only being taken after non-conforming products are produced, leading to waste of materials and time and failing to fundamentally guarantee the consistency and reliability of product batch quality. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a monitoring method for an antistatic foam box production line for liquid crystal displays, which solves the problem of lacking dynamic correlation and causal chain mining among parameters of multiple stages throughout the entire process.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring a production line of antistatic foam boxes for liquid crystal displays, comprising the following steps: Step S1: Collect historical operating data of each section in the production line of antistatic foam boxes for LCD displays, and perform data alignment and preprocessing to obtain historical production line data; Step S2: Use the time series forest algorithm to extract features from the historical data of the production line to obtain the dynamic time series features of the production line. Use the decision tree model to sort out the causal relationship of the dynamic time series features of the production line and generate the mutual intuition rule base of the production line. Step S3: Obtain the real-time dynamic time sequence characteristics of the production line, input the real-time dynamic time sequence characteristics of the production line into the production line mutual inductance rule base for matching, and locate the root cause section and parameters; Step S4: Based on the root cause section and parameters, simulate intervention on the shadow production line in the antistatic foam box production line. If the simulation intervention effect reaches the expected value, then intervene and adjust the main production line of the antistatic foam box production line. Through hierarchical model predictive control MPC algorithm, perform collaborative optimization to ensure that the main production line maintains stable operation under the intervention plan and produces batches of antistatic foam box products ready for release. Step S5: Perform electrostatic safety load verification on the batch of antistatic foam boxes to obtain the electrostatic safety load results.

[0006] Preferably, historical operating data of each section in the antistatic foam box production line for LCD displays is collected, including: In the five core sections of the antistatic foam box production line for LCD displays, namely the mixing section, pre-foaming section, molding section, cooling and shaping section, and surface processing section, dedicated sensor systems are deployed to collect historical operating data that can represent the key parameters of the section. From the complex operational data of each section, a set of key process parameters that are strongly correlated with the electrostatic safety state are precisely defined, which is called the parameter fingerprint set; To facilitate subsequent mathematical modeling and feature extraction, each parameter fingerprint is assigned a unique mathematical symbol, let the first... The parametric fingerprint vector of each work section is denoted as: ,in These correspond to five work sections in sequence.

[0007] Preferably, the historical data obtained from the production line includes: The raw data collected above, bearing a unified timestamp, undergoes preprocessing to generate high-quality historical production line data suitable for subsequent analysis. The preprocessing mainly includes data cleaning and time-series alignment. For data cleaning, outliers in sensor signals are identified and processed; for null values, a time-series-based linear interpolation algorithm is used to fill them in; for jump values ​​that clearly exceed physical meaning, the moving average of the preceding and following data is used for replacement. For timing alignment, a lightweight alignment method based on fixed delay compensation and linear interpolation is used to determine the production line section with the most stable cycle time and serving as the production trigger source as the benchmark section. The trigger signal of this section is the main cycle time signal. During the system initialization phase, the fixed transmission delay of sensor data from other sections relative to the main cycle time signal of the benchmark section is experimentally determined. subscript This represents the corresponding work section, and the original data timestamps of all work sections are compensated, i.e. Subsequently, using the main cycle time of the benchmark section as the time axis, a linear interpolation algorithm was used to resample the compensated data of each section onto a unified time sequence. The historical data of the production line for antistatic foam boxes, generated after the above preprocessing, can be represented by a set:

[0008] in, It is a batch code. It's a timestamp. It is a mathematical symbol that represents all; This means for all batches; This represents all time points within each batch.

[0009] Preferably, the historical data of the production line is used to extract features using a time-series forest algorithm to obtain dynamic time-series features of the production line, including: For any batch Any parameter fingerprint A time series, defined as a length of A sliding window is used to extract a set of statistical features within the window to form a dynamic time-series feature vector of the parameter within a specific time period of the batch. The extracted features include: mean Reflecting parameters Average level within the window; standard deviation Reflecting parameters The degree of fluctuation within the window; the slope of the trend. Obtained through linear fitting, reflecting the parameters The changing trend; approximate entropy Reflecting parameters The complexity and regularity of time series; definition Through this process, the raw data of each batch is transformed into a production line dynamic time series feature dataset composed of dynamic time series features of each section and parameter.

[0010] Preferably, the production line's dynamic temporal characteristics are analyzed causally using a decision tree model to generate a production line mutual inductance rule base, including: After obtaining the dynamic time-series feature dataset of the production line, the goal is to uncover the causal relationships between the features of different work sections. To this end, a highly interpretable decision tree model is used for analysis. Model training and rule extraction were performed using the dynamic temporal features of all work sections. Using the electrostatic safety performance index of the final product as the prediction target, the decision tree model is trained. After training, the path from the root node to each leaf node is extracted from the decision tree. Each path is a classification rule in the form of IF-THEN. Next, a production line mutual inductance rule base is constructed. These rules are organized, redundancies are removed, and priorities are sorted to form the production line mutual inductance rule base, denoted as . Each rule in this rule base Each contains three core components: triggering conditions, prediction results, and root cause localization. The triggering conditions are constituted by the dynamic temporal characteristics of one or more previous work sections meeting a specific threshold range; the prediction results indicate specific electrostatic safety problems that may occur in future work sections without intervention; and the root cause location clearly identifies the key root cause work section number that triggers this rule. and specific parameter fingerprint number This design directly serves S3, providing targets for generating precise intervention plans.

[0011] Preferably, the real-time dynamic time-series characteristics of the production line are obtained, and these characteristics are input into the production line mutual inductance rule base for matching to locate the root cause section and parameters, including: Data is collected in real time from sensors at each stage of the antistatic foam box production line and recorded as real-time production line data. The data acquisition source, synchronization mechanism, and preprocessing process are consistent with those in step S1. Subsequently, following the feature extraction method in step S2, dynamic temporal features are calculated on the real-time data stream. To distinguish it from the historical features in S2, the dynamic temporal feature vector extracted from the current real-time data is denoted as... The set of these features is called the real-time dynamic time series feature set. This step transforms continuous real-time signals into structured features that can be used for pattern matching by the rule base. Real-time dynamic time series feature set Input into the production line mutual inductance rule base generated by S2 Real-time matching is performed, which involves searching for all rules. , making Satisfy the rules The triggering condition, once matched successfully, will be determined according to the rules. The root cause location information defined in the code immediately identifies the critical root cause section number that triggers the abnormal trend. and parameters .

[0012] Preferably, based on the root cause section and parameters, the simulation intervention of the shadow production line in the antistatic foam box production line includes: Based on the root cause section located in step S3 and parameters Generate structured intervention plans Its generation principle is to map the abnormal state of the root cause parameter into a specific control command based on the adjustment strategy preset in the historical experience or process knowledge base. A complete intervention program It should include the following core elements: intervention objective, adjustment actions, window of action, and expected verification points: For the intervention target, clearly specify the work section that needs adjustment. and parametric fingerprints ; For adjustment actions, specify the specific operations, directions, and recommended ranges for parameter adjustments; for the action window: define the start time, duration, or end conditions of the intervention; for the expected verification points: specify which subsequent work section and which parameter to monitor to quickly assess the intervention effect; A shadow production line is set up that is completely identical to the main production line in terms of process equipment and control logic. It is used for safety verification of intervention schemes. The main production line is used for actual production, and the shadow production line is used for simulation testing.

[0013] Preferably, if the simulated intervention effect reaches the expected value, then the main production line of the antistatic foam box production line is intervened and adjusted, including: The generated intervention plan The process involves deploying a shadow production line for simulation verification. The verification process includes two core evaluation dimensions, and quantitative indicators are introduced for judgment, encompassing effectiveness assessment and impact assessment. For effectiveness evaluation, the main focus is on assessing the improvement effect of the intervention program on electrostatic safety and quality, defining the rate of decrease in quality risk indicators. for:

[0014] in, and These respectively represent the implementation of intervention plans on the shadow production line. Before and after, the average value of the key quality risk indicators monitored; when When the validity verification is passed, A validity threshold pre-set based on statistical analysis of historical data; For impact assessment, the main focus is on evaluating the impact of intervention programs on production stability and energy efficiency, defining production cycle fluctuation rate. and the rate of change of unit energy consumption for:

[0015] in, and These represent the implementation of intervention plans on the shadow production line. The average production cycle time obtained before and after; and These represent the implementation of the intervention plan. Before and after the intervention, the average energy consumption per unit of product was statistically obtained; the requirements for the intervention were as follows: and ,in and The preset tolerance threshold is based on production requirements. This is the maximum permissible production cycle time fluctuation, typically set based on the adjustable range of the production plan. It is the maximum permissible rate of change in unit energy consumption, which is usually set based on energy efficiency assessment indicators; An intervention plan is determined only if both of the above assessments are passed. Verification successful; migration to the mainline is permitted.

[0016] Preferably, collaborative optimization is achieved through hierarchical model predictive control (MPC) algorithm to ensure stable operation of the main production line under the intervention scheme, resulting in batches of antistatic foam boxes awaiting release, including: Once the intervention plan Or a new optimized solution Once validated on the shadow production line, it needs to be safely and smoothly migrated to the main production line for execution. A hierarchical predictive control architecture is adopted. Specifically, the operation flow of this hierarchical control system includes upper-level controllers and lower-level controllers: The upper-level controller is responsible for strategic rules on a slow timescale; In each optimization cycle, the upper-level controller, based on the overall macro model of the production line and the current state, solves for two key decision variables: the global production cycle time setpoint and the global production takt time setpoint. The first purpose is to coordinate the production rates of each work section and avoid bottlenecks; the second is to allocate energy consumption budgets for each work section. For each work section Set upper limits for energy consumption to achieve energy efficiency management; For the objective function, its optimization objective is to satisfy the intervention plan. Given the requirements, and taking into account both production efficiency and energy consumption, this objective can be formally expressed as a weighted summation function:

[0017] in, This is a cost term characterizing total energy consumption, defined as the estimated total energy consumption of the main production line during the intervention period, calculated using the following formula: , Representing the Estimated energy consumption of each work section during the intervention period; It is a cost term characterizing production capacity, defined as the negative value of total output during the intervention period, i.e. , This represents the expected number of production batches to be completed within this period. Minimizing this cost term is equivalent to maximizing total output; weighting coefficients. and This is pre-set according to the actual production strategy. When production tasks are tight and priority needs to be given to ensuring delivery, the following settings are configured: The optimization objective will tend to maximize output; when the factory is in an energy-saving and consumption-reducing cycle, then the following settings will be implemented: The optimization objective will tend to minimize energy consumption; the output of the upper-level controller is the globally optimal setpoint combination. ; The lower-level controller is responsible for tactical execution on a fast timescale: Each section is equipped with an independent MPC controller, which operates rapidly at a cycle of seconds or less to ensure timely response to process disturbances and setpoint changes. The core task of each section controller is to drive the process parameters of its section to track the set values ​​issued by the upper level quickly and accurately, while maintaining smooth control actions and avoiding impact on the equipment. The optimization problem of the MPC controller at each work section level is solved within each control cycle, and its objective function is typically designed as follows:

[0018] in, For the prediction time domain, it is a preset positive integer representing how far into the future the controller will predict the system output; To control the time domain, it is a preset positive integer that satisfies... This indicates how far in the future the controller will optimize the control increments; It is the output variable vector of the work section; Is the section in the future Predicted values ​​of the output variable vector at time step; It is the corresponding setting value issued by the upper-level MPC. It manipulates the variable vector; Indicates the change in the control quantity; In the future Control increment at any given moment; and The weight matrix penalizes both the output tracking error and drastic changes in the control quantity, and needs to be determined during on-site commissioning based on the dynamic characteristics of each section; index Used to iterate through future moments within the prediction time domain; index , used to traverse the control steps within the control time domain; Through closed-loop interaction between upper and lower level controllers, dynamic collaboration is achieved. When an intervention plan causes a change in the setpoint of a certain section, the upper-level MPC will recalculate the globally optimal cycle time and energy consumption allocation, and issue new instructions to all sections. The lower-level MPC controllers will immediately respond, adjusting their local actuators to ensure that the output of their section tracks the new setpoint. Through this mechanism of upper-level decision-making and lower-level execution, the entire production line can smoothly and stably transition to the new operating conditions after intervention, and continuously maintain a dynamic balance between production cycle time and energy consumption while meeting quality requirements, producing batches of antistatic foam boxes ready for release. .

[0019] Preferably, the electrostatic safety load of the batch of antistatic foam boxes is verified, and the electrostatic safety load results include: From batch Representative samples were selected and placed in a simulated test environment that could simulate specific relative humidity, vibration frequency, and friction conditions. A pulse of a specified voltage was applied using a standardized electrostatic discharge device, and the charge decay time on the surface of the enclosure was immediately measured. and residual voltage ; Measurement results Safety thresholds specified in the product specifications Compare; if satisfied and If the electrostatic safety verification is passed, the batch is deemed to have passed and is permitted to be released.

[0020] This invention provides a method for monitoring a production line of antistatic foam boxes for liquid crystal displays, involving machine learning and deep learning technologies, which has the following beneficial effects: (1) This monitoring method for the production line of antistatic foam boxes for LCD displays improves upon traditional knowledge acquisition methods by employing a decision tree algorithm to automatically mine and generate IF-THEN causal rules from the dynamic temporal characteristics of the production line. Compared to relying on expert experience for manual summarization or using black-box machine learning models, this improvement makes the constructed production line mutual inductance rule base highly interpretable. Each rule clearly reveals the causal path between abnormal parameters in a specific section and the final quality defect. This allows engineering technicians to intuitively understand, verify, and trust the system's diagnostic logic, facilitating rapid on-site location of the root cause of the problem and providing a clear basis for manual review and rule optimization, greatly enhancing the practicality and acceptability of the monitoring system.

[0021] (2) The monitoring method for the antistatic foam box production line of the LCD display screen has made key improvements to the traditional feature extraction method. It adopts the Time Series Forest (TSF) algorithm to extract dynamic time-series features from the original parameter sequences of each section. Compared with the traditional method, which only focuses on the instantaneous value of the parameter or simple statistics, this improvement can more comprehensively and deeply characterize the dynamic behavior and evolution pattern of the production process parameters by calculating higher-order statistical features such as trend slope, fluctuation entropy, and approximate entropy within the sliding window.

[0022] (3) Monitoring method for antistatic foam box production line of LCD display screen: By applying model predictive control algorithm to the antistatic foam box production line, traditional single-loop control cannot handle this coupling, which easily leads to system oscillation where pressing down the gourd causes the ladle to float. The core advantage of MPC is its multivariable control capability. This method processes all relevant input and output variables in a unified optimization problem, automatically coordinates the actions between various manipulated variables, and thus perfectly solves the coupling problem between variables, realizing the overall line collaborative optimization.

[0023] (4) Monitoring method for antistatic foam box production line of LCD display screen: Through structural innovation and improvement of the traditional model predictive control application method, a hierarchical predictive control architecture was designed. This improvement decomposes the complex whole-line optimization control problem into two levels: the upper scheduling layer MPC performs slow-cycle global strategic optimization and sets the whole-line cycle time and energy consumption allocation; the lower section layer MPC performs fast-cycle local tactical tracking control. This improvement effectively solves the curse of dimensionality problem faced by a single MPC controller when dealing with large-scale, multi-variable, and dynamically diverse production lines, and breaks through the bottleneck of traditional control methods that cannot simultaneously take into account global optimization and local dynamic performance. Attached Figure Description

[0024] Figure 1 This is a flowchart of a monitoring method for an anti-static foam box production line for liquid crystal displays, as proposed in this invention.

[0025] Figure 2 The method for monitoring a production line of antistatic foam boxes for liquid crystal displays proposed in this invention obtains a hierarchical diagram of the root cause sections and parameters.

[0026] Figure 3 This is a hierarchical diagram showing the electrostatic safety load results obtained in the monitoring method for an antistatic foam box production line of a liquid crystal display screen proposed in this invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1-3 This invention provides a technical solution: a method for monitoring a production line of antistatic foam boxes for liquid crystal displays. Specifically, the method for monitoring a production line of antistatic foam boxes for liquid crystal displays is provided below. Please refer to [link / reference]. Figure 1 The method includes the following steps: Step S1: Collect historical operating data of each section in the production line of antistatic foam boxes for LCD displays, and perform data alignment and preprocessing to obtain historical production line data.

[0029] First, dedicated sensor systems are deployed in the five core sections of the antistatic foam box production line for LCD displays, namely the mixing section, pre-foaming section, molding section, cooling and shaping section, and surface processing section, to collect historical operating data of key parameters that represent the section. Due to the large amount of historical operating data of the production line, in order to effectively monitor the core quality attribute of the product—electrostatic safety, this method first precisely defines a set of key process parameters that are strongly related to the electrostatic safety state from the complex operating data of each section, which is called the parameter fingerprint set. To facilitate subsequent mathematical modeling and feature extraction, each parameter fingerprint is assigned a unique mathematical symbol. Let the... The parametric fingerprint vector of each work section is denoted as: ,in These correspond to five work sections in sequence. Two key parameter fingerprints are collected for each work section, defined as follows: In the mixing section (k=1), the first parameter fingerprint collected is defined as the raw material moisture content trend. It obtains the average moisture content of the material and its slope over time by monitoring it with an online near-infrared sensor; the second parameter fingerprint is defined as the filler wetting adequacy. It indirectly reflects the uniformity of conductive filler dispersion in the matrix by monitoring the torque current change of the main motor of the mixing machine; the parametric fingerprint vector of this section is ; In the pre-foaming section (k=2), the first parametric fingerprint collected is defined as bubble growth stability. It characterizes the distribution variance of bubble diameter in real time through an online visual inspection system; the second parameter fingerprint is defined as wall thickness uniformity. It obtains the parametric fingerprint vector by measuring the wall thickness of the foam at different points using an ultrasonic thickness gauge and calculating its standard deviation; the parametric fingerprint vector for this section is... ; In the molding process (k=3), the first parametric fingerprint collected is defined as the mold fit integrity. It is characterized by the pressure distribution uniformity coefficient of the distributed pressure sensor array within the mold cavity; the second parameter fingerprint is defined as the continuity of the conductive path formation. It is obtained by monitoring the volume resistivity of the material inside the mold cavity under low pressure using an online impedance analyzer; the parametric fingerprint vector of this section is ; In the cooling and shaping section (k=4), the first parametric fingerprint collected is defined as shrinkage uniformity. It characterizes the deformation consistency of key dimensions of the product during the cooling process by measuring the deformation consistency during the cooling process using a laser displacement sensor; the second parameter fingerprint is defined as the degree of internal stress release. It is obtained by online monitoring of the decay rate of residual stress inside the product using a polarized optical stress meter; the parametric fingerprint vector of this section is ; In the surface application section (k=5), the first parametric fingerprint collected is defined as the coating density. The areal density of the antistatic coating was obtained by measuring it with an online beta-ray thickness gauge; the second parameter fingerprint was defined as the surface charge decay performance. It characterizes the process by measuring the time required for the surface charge of a workpiece to decay to half its initial value after a standard charge is applied using a non-contact electrometer; the parametric fingerprint vector for this section is... .

[0030] The historical data of the antistatic foam box production line should come from the dataset collected by the aforementioned sensors under normal working conditions, rather than being randomly extracted from the historical dataset afterward. This is to ensure that the original characteristics of the source data strictly correspond to the production cycle. To overcome the timing discrepancies caused by independent clocks in different production sections, establishing a unified time reference is crucial. Specifically, a high-precision clock source is set up in the central control system of the production line to generate a timing reference signal that runs throughout the entire line. Each production batch is recorded as N represents the total number of production batches recorded on the production line; within that batch, the number of batches collected according to the timing reference signal is the first... Each time point is recorded as All sensor data acquisition triggers in all work sections are synchronized with this signal, and a uniform timestamp is added to each acquired data packet. and batch code .

[0031] To overcome the timing discrepancies caused by independent clocks in different production sections, establishing a unified time reference is crucial. Specifically, a high-precision clock source is set up in the central control system of the production line to generate a timing reference signal that runs throughout the entire line. Each production batch is recorded as N represents the total number of production batches recorded on the production line. Within that batch, the [number]th [batch] is the [number]th [batch] acquired based on the timing reference signal. Each time point is recorded as , This represents the j-th production batch (i.e. Within this timeframe, the total number of valid time points collected based on a unified timing reference signal. All data acquisition triggers of sensors in all work sections are synchronized with this signal, and each acquired data packet is stamped with a unified timestamp. and batch code .

[0032] Subsequently, the raw data with uniform timestamps collected above were preprocessed to form high-quality historical production line data that can be used for subsequent analysis. The preprocessing process mainly includes data cleaning and time-series alignment: For data cleaning, outliers in sensor signals are identified and processed, such as null values ​​caused by signal loss or transient jumps caused by interference. For null values, a time-series-based linear interpolation algorithm is used to fill them in; for jump values ​​that are clearly beyond physical meaning, the moving average of the preceding and following data is used for replacement.

[0033] For timing alignment, despite a unified timing benchmark, slight delays may exist in data transmission between different sensors. To address this issue, a lightweight alignment method based on fixed delay compensation and linear interpolation is employed. The section on the production line with the most stable cycle time, serving as the production trigger source, is designated as the benchmark section. The trigger signal of this section becomes the main cycle time signal. During system initialization, the fixed transmission delay of sensor data from other sections relative to the main cycle time signal of the benchmark section is experimentally determined. subscript This represents the corresponding work section, and the original data timestamps of all work sections are compensated, i.e. Subsequently, using the main cycle time of the benchmark section as the time axis, a linear interpolation algorithm was employed to resample the compensated data of each section onto a unified time sequence.

[0034] Finally, the historical data of the production line for antistatic foam boxes, after the above preprocessing, can be represented by the following set:

[0035] in, It is a batch code. It's a timestamp. It is a mathematical symbol that represents all; This indicates that for all batches (j ranges from 1 to the total batch N); This represents all time points within each batch. Each unit group This defines a basic data record in the set.

[0036] It should be noted that the data in this step comes from the historical operating data of the main production line. This step is the data foundation stage, and its core task is to accurately define and collect key parameters, i.e., parameter fingerprint sets, that represent the electrostatic safety state of each of the five core processes: mixing, pre-foaming, molding, cooling, and surface application. By deploying dedicated sensors and establishing a unified time-series reference signal across the entire line, the time synchronization problem of multi-source data is solved. Subsequently, the collected raw data is cleaned and preprocessed with time-series alignment based on fixed delay compensation, ultimately generating a high-quality, time-consistent, and analyzable historical dataset of the production line.

[0037] Step S2: Use the time series forest algorithm to extract features from the historical data of the production line to obtain the dynamic time series features of the production line. Use the decision tree model to analyze the causal relationship of the dynamic time series features of the production line and generate a mutual intuition rule base for the production line.

[0038] First, from the historical production line data obtained by S1 Parametric fingerprint time series of each work section Deep feature mining is performed, where n is the index of the parametric fingerprint. Directly using the original instantaneous values ​​is insufficient to capture the dynamic evolution of the process. Therefore, this step borrows the core idea of ​​the Time Series Forest (TSF) algorithm to extract features representing the dynamic behavior of the time series within a sliding window.

[0039] Specifically, for any batch Any parameter fingerprint A time series, defined as a length of The sliding window, based on prior knowledge This can be set to the number of data points corresponding to the time required for the main equipment (such as a mixer or molding press) of this section to complete one basic operation unit. For example, if the typical cycle for the mixing section to complete one material mixing operation is... seconds, data sampling interval is If the window length is 1 second, then the window length can be initialized to 1. The step size can be set to an integer multiple of the window length. A fraction, for example or This creates a gap between adjacent windows. or The overlap rate ensures continuous monitoring of the process status, so that even if a momentary anomaly does not fall completely in the center of a window, it can be effectively captured by adjacent windows. Within this window, a set of statistical features is extracted to form a dynamic time-series feature vector of this parameter within a specific time period of this batch. The extracted features include: mean Reflecting parameters Average level within the window; standard deviation Reflecting parameters The degree of fluctuation within the window; the slope of the trend. Obtained through linear fitting, reflecting the parameters The changing trend; approximate entropy Reflecting parameters The complexity and regularity of time series.

[0040] Therefore, the definition Through this process, the raw data for each batch is transformed into a dynamic time-series feature dataset for the production line, consisting of dynamic time-series features of each work section and parameter. This dataset provides a more comprehensive characterization of the dynamic behavior of the production process.

[0041] After obtaining the dynamic time-series feature dataset of the production line, the goal is to uncover the causal relationships between the features of different work sections. To this end, a highly interpretable decision tree model is used for analysis. First, model training and rule extraction are performed, using the dynamic time-series features of all work sections. Using the electrostatic safety performance index of the final product as the prediction target, the decision tree model is trained. After training, the path from the root node to each leaf node is extracted from the decision tree. Each path is a classification rule in the form of IF-THEN. IF means if, followed by the premise; THEN means then, followed by the conclusion. The IF-THEN form specifically refers to the form adopted by each rule in the mutual inductance rule base of the production line.

[0042] Next, a production line mutual inductance rule base is constructed. These rules are organized, redundancies are removed, and priorities are sorted to form the production line mutual inductance rule base, denoted as . Each rule in this rule base Each contains three core components: triggering conditions, prediction results, and root cause localization. The triggering conditions are constituted by the dynamic temporal characteristics of one or more previous work sections meeting a specific threshold range; the prediction results indicate specific electrostatic safety problems that may occur in future work sections without intervention; and the root cause location clearly identifies the key root cause work section number that triggers this rule. and specific parameter fingerprint number This design directly serves S3, providing targets for generating precise intervention plans.

[0043] Suppose a rule generated by the decision tree model As shown below: The trigger condition is IF mixing section ( The parameter of fingerprint filling adequacy ( The dynamic characteristic trend slope (i.e., showing a significant downward trend) AND pre-foaming section ( The parameter fingerprint bubble growth stability ( The dynamic characteristic standard deviation (i.e., abnormally large fluctuations); The prediction result is THEN's prediction in the compression molding section ( The risk level of the formation of discontinuous conductive pathways is high.

[0044] Regarding the root cause, it lies in the mixing section ( ) The degree of wetting of the filler ( The trend of deterioration.

[0045] To ensure the reliability of the rule base, the generated rules are validated. This is achieved by analyzing the consistency of actual quality results after rules are triggered in historical data, and calculating the validity of each rule. confidence level :

[0046] in, Indicates support for the rule The batch quantity, i.e., the rule The number of batches for which the triggering condition is met and the prediction result is true; Representation rules The total number of batches triggered, i.e., the rule The number of batches whose triggering conditions are met. This formula quantifies the rule. The reliability of a rule is determined by the conditional probability that the predicted result will actually occur when the rule's triggering conditions are met. A higher confidence level indicates a more reliable rule.

[0047] In actual screening, only when At that time, the rules Only then will it be retained in the rule base. In the middle, only those with confidence levels higher than a preset threshold are retained. Strong rules are entered into the database to ensure the accuracy and actionability of the rule base. The specific value is a threshold value pre-set based on engineering experience and historical data verification results. It can be set... This means that a rule is considered reliable only if its prediction accuracy in historical data is higher than 85%.

[0048] The core task of this step is to mine potential patterns across work sections from the historical data of the S1 production line. First, drawing on the concept of Time Series Forest (TSF), the parametric fingerprint time series of each work section are transformed into dynamic time-series features that better reflect process dynamics. Then, a decision tree model is used to train and learn from these dynamic features, automatically identifying causal chains between work sections, and refining these causal relationships into interpretable IF-THEN rules, ultimately generating a production line mutual intuition rule base. This rule base also includes a confidence assessment mechanism to ensure the reliability of the knowledge.

[0049] Step S3: Obtain the real-time dynamic time sequence characteristics of the production line, input the real-time dynamic time sequence characteristics of the production line into the production line mutual inductance rule base for matching, and locate the root cause section and parameters.

[0050] First, real-time data is collected from sensors at each stage of the antistatic foam box production line and recorded as real-time production line data. The data acquisition source, synchronization mechanism, and preprocessing procedure remain consistent with S1. Then, following the feature extraction method in step S2, dynamic temporal feature calculations are performed on the real-time data stream. To distinguish it from the historical features in S2, the dynamic temporal feature vector extracted from the current real-time data is denoted as... The set of these features is called the real-time dynamic time series feature set. This step transforms continuous real-time signals into structured features that can be used for pattern matching by the rule base.

[0051] Real-time dynamic time series feature set Input into the production line mutual inductance rule base generated by S2 Real-time matching is performed, which involves searching for all rules. , making Satisfy the rules The triggering condition. Once a match is successful, it is determined according to the rules. The root cause location information defined in the code immediately identifies the critical root cause section number that triggers the abnormal trend. And parameters (parameter fingerprint number) .

[0052] For example, suppose there are rules in the rule base. Its definition is: The trigger condition is IF (That is, the slope of the trend of packing wettability in the mixing section is less than -0.5, showing a rapid decrease) AND (That is, the standard deviation of bubble growth stability in the pre-foaming section is greater than 2.0, and the fluctuation is drastic); The prediction result indicates that the risk level of discontinuous conductive path formation in the THEN molding section is high. The root cause lies in the [mixing section ( ), parameter packing wetting adequacy ( )).

[0053] If real-time dynamic time series feature set If the calculation result satisfies the above triggering conditions, then the rule Upon being triggered, the root cause of the problem was immediately diagnosed as the mixing section. The second parameter fingerprint () ), which refers to the degree of wetting of the filler.

[0054] It should be noted that this step is the real-time monitoring phase, and the monitoring object is the real-time data of the main production line. Rule base matching and root cause identification are also based on the current state of the main production line.

[0055] This step serves as the real-time diagnostic and decision-making hub in the monitoring process. Its core task is to apply the historical patterns extracted from S2 to online production, achieving a leap from monitoring to diagnosis and then to decision-making. By collecting production line data in real time and extracting dynamic time-series features, and then quickly matching them with a pre-generated mutual intuition rule base, the system can accurately locate the root cause of abnormal trends and specific parameters before product quality defects actually occur.

[0056] Step S4: Based on the root cause section and parameters, simulate intervention on the shadow production line in the antistatic foam box production line. If the simulation intervention effect reaches the expected value, then intervene and adjust the main production line of the antistatic foam box production line. Through hierarchical model predictive control MPC algorithm, perform collaborative optimization to ensure that the main production line maintains stable operation under the intervention plan and produces batches of antistatic foam box products ready for release.

[0057] Based on the root cause section located in step S3 and parameters Generate structured intervention plans Its generation principle is to map the abnormal state of the root cause parameter into specific control commands based on the adjustment strategy preset in the historical experience or process knowledge base.

[0058] A complete intervention program It should include the following core elements: intervention objective, adjustment actions, window of action, and expected verification points: For the intervention target, clearly specify the work section that needs to be adjusted. ) and parametric fingerprints ( ); For adjustment actions, specify the specific operation, direction, and recommended range of parameter adjustment; for the action window: define the start time, duration, or end conditions of the intervention; for the expected verification point: specify which subsequent work section and which parameter to monitor to quickly evaluate the intervention effect.

[0059] Continuing from the previous example, regarding the rules The root cause of the trigger, and the resulting intervention plan This can be specified as: The intervention target is the mixing section ( ), Parametric fingerprint filler wetting adequacy ( ); The adjustment action is to increase the setpoint of the main motor speed of the mixing mill. And increase the temperature setpoint of the mixing zone. The action window is for immediate intervention, and the duration is [duration missing]. Minutes; the expected verification point is to monitor the bubble growth stability of the subsequent pre-foaming section. Whether the fluctuations of ( ) tend to stabilize.

[0060] This method sets up a shadow production line that is completely consistent with the process equipment and control logic of the main production line for the safety verification of the intervention scheme. The main production line is used for actual production, and the shadow production line is used for simulation testing. The generated intervention plan The process involves deploying a shadow production line for simulation verification. The verification process includes two core evaluation dimensions, and quantitative indicators are introduced for judgment, encompassing effectiveness assessment and impact assessment. For effectiveness evaluation, the main focus is on assessing the intervention's improvement effect on electrostatic safety and quality. The core function of the antistatic foam box is to safely dissipate static charge through a network path formed by internal conductive filler, preventing damage to the LCD screen. Volume resistivity is a physical quantity that measures a material's conductivity; the lower the resistivity, the more continuous the conductive path and the stronger the static dissipation capacity. The rate of decrease in quality risk indicators is defined as... for:

[0061] in, and These respectively represent the implementation of intervention plans on the shadow production line. Before and after, the average value of the key quality risk indicators monitored (such as the predicted risk value of the discontinuity of the conductive path in the molding section). When When the validity verification is passed, it is considered to be valid. This is a pre-set validity threshold based on statistical analysis of historical data.

[0062] It should be noted that the intervention plan for the shadow production line... Previously, continuous collection One production cycle (e.g.) Volume resistivity data , That is this The arithmetic mean of the data points is calculated using the following formula:

[0063] in, It is an intervention plan implemented on the shadow production line. The number of pre-collected production cycles, It is the second parameter of the fingerprint resistivity data in the molding stage, and i is the index of the periodic sequence; Post-intervention effect value In intervention program After the shadow production line takes effect and runs stably, data will be collected continuously. One production cycle (e.g.) The volume resistivity data of ) That is this The arithmetic mean of the data points is calculated using the following formula:

[0064] in, It is an intervention plan implemented on the shadow production line. The number of subsequent production cycles collected; For impact assessments, the primary focus is on evaluating the impact of interventions on production stability and energy efficiency. Production cycle time volatility is defined. and the rate of change of unit energy consumption for:

[0065] in, and These represent the implementation of intervention plans on the shadow production line. The average production cycle time obtained before and after; and These represent the implementation of the intervention plan. Before and after the intervention, the average energy consumption per unit of product was statistically obtained; the requirements for the intervention were as follows: and ,in and The preset tolerance threshold is based on production requirements. This is the maximum permissible production cycle fluctuation rate, typically set based on the adjustable range of the production plan (e.g., 2%). It is the maximum allowable rate of change in unit energy consumption, which is usually set based on energy efficiency assessment indicators (such as 3%).

[0066] An intervention plan is determined only if both of the above assessments are passed. Verification successful; migration to the mainline is permitted.

[0067] If intervention plan If the simulation results in the shadow production line do not meet expectations, the scheme optimization mechanism will be activated: First, analyze the main reasons for not meeting the standards, such as insufficient quality improvement or unacceptable disturbances to production cycle or energy consumption; The iterative optimization process for the proposed solution is as follows: Based on the analysis results, the intervention plan is adjusted according to the preset optimization algorithm (such as gradient descent, genetic algorithm, etc.). Fine-tune the adjustments made in the process (such as the temperature adjustment range and the speed change) to generate an optimized new solution. ; If simple parameter fine-tuning is ineffective, the failed case is fed back to the rule mining process of S2 to check whether the confidence of the triggering rule needs to be lowered, or whether there are more complex coupling relationships that are not covered by the existing rule base.

[0068] The optimized new solution The system re-enters the shadow production line for simulation verification. If the optimized solution passes the verification within the preset number of iterations (e.g., 3 times), the system will pass the verification. If the system fails to pass the verification after exceeding the maximum number of iterations, the system will generate a high-level alarm and push the root cause analysis report, all attempted intervention solutions, and simulation results to the engineer, requesting manual intervention for decision-making.

[0069] Once the intervention plan Or a new optimized solution Once validated on the shadow production line, it needs to be safely and smoothly migrated to the main production line. However, adjusting parameters in a single section can have a domino effect, impacting upstream and downstream sections through the coupling of material and energy flows, disrupting the overall line balance. To address this multivariate collaborative control challenge, this method employs a hierarchical model predictive control (Hierarchical MPC) architecture. The core idea of ​​this architecture is to decompose the complex line control problem according to time scale and decision level, achieving a unification of global optimization and local dynamic control through the collaborative operation of upper and lower level controllers.

[0070] Specifically, the operation process of this hierarchical control system is as follows: First, the upper-level controller (scheduling layer MPC) is responsible for strategic planning on slow timescales; In each optimization cycle, the upper-level controller, based on the overall macro model of the production line and the current state, solves for two key decision variables: the global production cycle time setpoint and the global production takt time setpoint. The first purpose is to coordinate the production rates of each work section and avoid bottlenecks; the second is to allocate energy consumption budgets for each work section. For each work section Set upper limits for energy consumption to achieve energy efficiency management; For the objective function, its optimization objective is to satisfy the intervention plan. Under the given requirements (i.e., quality constraints), and taking into account both production efficiency and energy consumption, this objective can be formally expressed as a weighted summation function:

[0071] in, It is a cost term characterizing the total energy consumption. It is a cost term representing production capacity, with a weighting coefficient. and It is then pre-set according to the actual production strategy to weigh the importance between energy efficiency and output.

[0072] The output of the upper-level controller is the globally optimal setpoint combination. .

[0073] Then, the lower-level controller (section-level MPC) is responsible for tactical execution on a fast timescale: Each section is equipped with an independent MPC controller, which operates rapidly at a cycle of seconds or less to ensure timely response to process disturbances and setpoint changes. The core task of each section controller is to drive the process parameters (such as mixing temperature and molding pressure) of its section to quickly and accurately track the set values ​​issued from the upper level, while maintaining smooth control actions to avoid impacting the equipment.

[0074] To achieve the above objectives, the optimization problem of the MPC controller at each work section level is solved within each control cycle, and its objective function is typically designed as follows:

[0075] in, To predict the time domain, To control the time domain. It is the output variable vector of the work section (such as temperature, pressure). It is a setting value issued by the higher-ups. It is a vector of manipulated variables (such as heater power, valve opening). It indicates the amount of change in the control quantity. and The weight matrix is ​​used to penalize the output tracking error and drastic changes in the control quantity, respectively. It needs to be determined during on-site debugging based on the dynamic characteristics of each section. The objective function aims to minimize the deviation between the output and the set value over a future period of time, while constraining the amplitude of the control action.

[0076] Finally, through closed-loop interaction between upper and lower level controllers, dynamic coordination is achieved. When an intervention plan causes a change in the setpoint of a certain section (such as the mixing section), the upper-level MPC will recalculate the globally optimal cycle time and energy consumption allocation, and issue new instructions to all sections. The lower-level MPC controllers will immediately respond, adjusting their local actuators to ensure that the output of their respective sections tracks the new setpoint. Through this mechanism of upper-level decision-making and lower-level execution, the entire production line can smoothly and stably transition to the new operating conditions after intervention, and, while meeting quality requirements, continuously maintain a dynamic balance between production cycle time and energy consumption, producing batches of antistatic foam boxes ready for release. .

[0077] Step S5: Perform electrostatic safety load verification on the batch of antistatic foam boxes to obtain the electrostatic safety load results.

[0078] For product batches that have been manufactured and are ready for release under the new operating conditions in S4 Electrostatic discharge (ESD) safety load verification is required. This verification aims to simulate the ESD risks that the product may encounter during actual transportation, warehousing, and handling. From batch Representative samples were selected and placed in a simulated testing environment. Since LCD screens are more susceptible to electrostatic discharge (ESD) in low humidity conditions, and high-frequency vibrations during transport may exacerbate wear on the foam box coating, affecting ESD protection, the simulated testing environment was based on typical LCD screen transport conditions, such as relative humidity controlled at 30%–60% (LCD screens are more sensitive to low humidity), increasing the vibration frequency to a higher frequency range (50–200 Hz) to simulate logistical bumps, and using specific materials (such as nylon or polyester) to simulate contact between the packaging and the equipment for friction conditions. Next, a pulse of a specified voltage is applied using standardized electrostatic discharge equipment, with the applied voltage range extended to 500V~2000V (higher than the general standard of 200V~1000V) to cover the sensitive threshold of the LCD screen. The charge decay time on the surface of the enclosure is then immediately measured. and residual voltage .

[0079] Measurement results The safety threshold specified in the product specifications of the antistatic foam box for LCD screens Compare, if satisfied and If the electrostatic safety verification is passed, the batch is deemed to have passed and is permitted to be released.

[0080] The results of load verification not only form the basis for batch release but also serve as crucial data driving the evolution of the system's knowledge base. This process establishes a closed loop of continuous learning: Batch The verification results are associated with their corresponding end-to-end data, including The original parameter fingerprint in The dynamic temporal features extracted from them Rules triggered in as well as Intervention programs implemented in China This forms a complete process-intervention-result data package.

[0081] And for the rules that were triggered during this production run. Based on the actual performance of this verification, the confidence level is dynamically adjusted using Bayesian update or its statistical methods. If a rule successfully predicts a quality problem or its intervention is effective, its confidence level is increased; otherwise, it is decreased. If a problem occurs in this production run that is not covered by the existing rule base... If an abnormal pattern is covered, but potential risks are discovered through load validation, then this process-outcome case is used as a new sample to trigger... The rule mining process generates new candidate rules. After verification using historical data, new rules with high confidence are added to the rule base. .

[0082] Similarly, based on the results of this verification, for Intervention plans generated in The effectiveness of the optimization scheme is evaluated, and parameters such as the range of motion and the time window of action are optimized.

[0083] This step is the closed-loop optimization phase. Its core task is to perform closed-loop quality verification of the final product produced under S4 collaborative control and drive the system's self-evolution. Before batch release, the product undergoes electrostatic safety load verification to simulate real-world conditions and test its performance. More importantly, the verification results are correlated with the entire process data from S1 to S4 to form a complete data package of process-intervention-result. Using this data package, the confidence levels of rules in the S2 rule base are updated or expanded, and the intervention strategies of S3 are optimized, thereby achieving continuous learning of the rule base.

[0084] This invention provides a monitoring method for an antistatic foam box production line for LCD displays, aiming to solve problems such as monitoring lag, isolated parameters, and energy consumption imbalance in traditional methods. This invention establishes a full-process intelligent monitoring system centered on the mutual inductance relationship of process parameters. By constructing a multi-stage parameter fingerprint and time-series consistency mapping, it automatically identifies the causal chain between stages such as mixing, foaming, molding, cooling, and surface application, realizing the linkage analysis and self-learning optimization between parameters. This method overcomes the limitations of traditional single-point control, possessing feedforward intervention capabilities for abnormal trends. It can automatically generate and verify intervention plans before product defects occur, realizing a shift from post-event correction to pre-event prevention. Simultaneously, the system establishes a collaborative control mechanism for energy consumption and electrostatic protection. While implementing electrostatic risk intervention, it automatically adjusts the production cycle based on energy consumption distribution and stage load status, achieving dual optimization of antistatic performance and energy utilization efficiency. Furthermore, through electrostatic safety load verification and on-site sampling inspection before batch release, the system not only ensures product quality but also automatically writes detection deviations back to the rule base, achieving self-evolution and long-term stable control. In summary, this method, through multi-layered technological innovations such as mutual inductance analysis, feedforward intervention, cycle time self-adjustment, and knowledge backwriting, constructs an interpretable, predictable, and sustainably optimized monitoring system for antistatic foam box production lines, significantly improving product consistency and production efficiency.

[0085] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 a process, method, article, or apparatus. Without further limitations, an element defined by the statement "including a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. A method for monitoring a production line of antistatic foam boxes for liquid crystal displays, characterized in that, Includes the following steps: Step S1: Collect historical operating data of each section in the production line of antistatic foam boxes for LCD displays, and perform data alignment and preprocessing to obtain historical production line data; Step S2: Use the time series forest algorithm to extract features from the historical data of the production line to obtain the dynamic time series features of the production line. Use the decision tree model to sort out the causal relationship of the dynamic time series features of the production line and generate the mutual intuition rule base of the production line. Step S3: Obtain the real-time dynamic time sequence characteristics of the production line, input the real-time dynamic time sequence characteristics of the production line into the production line mutual inductance rule base for matching, and locate the root cause section and parameters; Step S4: Based on the root cause section and parameters, simulate intervention on the shadow production line in the antistatic foam box production line. If the simulation intervention effect reaches the expected value, then intervene and adjust the main production line of the antistatic foam box production line. Through hierarchical model predictive control MPC algorithm, perform collaborative optimization to ensure that the main production line maintains stable operation under the intervention plan and produces batches of antistatic foam box products ready for release. Step S5: Perform electrostatic safety load verification on the batch of antistatic foam boxes to obtain the electrostatic safety load results.

2. The monitoring method for an anti-static foam box production line for liquid crystal displays according to claim 1, characterized in that, Collect historical operating data for each section of the antistatic foam box production line for LCD displays, including: In the five core sections of the antistatic foam box production line for LCD displays, namely the mixing section, pre-foaming section, molding section, cooling and shaping section, and surface processing section, dedicated sensor systems are deployed to collect historical operating data that can represent the key parameters of the section. From the complex operational data of each section, a set of key process parameters that are strongly correlated with the electrostatic safety state are precisely defined, which is called the parameter fingerprint set; To facilitate subsequent mathematical modeling and feature extraction, each parameter fingerprint is assigned a unique mathematical symbol, let the first... The parametric fingerprint vector of each work section is denoted as: ,in These correspond to five work sections in sequence.

3. The monitoring method for an anti-static foam box production line for a liquid crystal display screen according to claim 2, characterized in that, Obtain historical production line data, including: The raw data collected above, bearing a unified timestamp, undergoes preprocessing to generate high-quality historical production line data suitable for subsequent analysis. The preprocessing mainly includes data cleaning and time-series alignment. For data cleaning, outliers in sensor signals are identified and processed; for null values, a time-series-based linear interpolation algorithm is used to fill them in; for jump values ​​that clearly exceed physical meaning, the moving average of the preceding and following data is used for replacement. For timing alignment, a lightweight alignment method based on fixed delay compensation and linear interpolation is used to determine the production line section with the most stable cycle time and serving as the production trigger source as the benchmark section. The trigger signal of this section is the main cycle time signal. During the system initialization phase, the fixed transmission delay of sensor data from other sections relative to the main cycle time signal of the benchmark section is experimentally determined. subscript This represents the corresponding work section, and the original data timestamps of all work sections are compensated, i.e. Subsequently, using the main cycle time of the benchmark section as the time axis, a linear interpolation algorithm was used to resample the compensated data of each section onto a unified time sequence. The historical data of the production line for antistatic foam boxes, generated after the above preprocessing, can be represented by a set: ； in, It is a batch code. It's a timestamp. It is a mathematical symbol that represents all; This means for all batches; This represents all time points within each batch.

4. The monitoring method for an anti-static foam box production line for a liquid crystal display screen according to claim 3, characterized in that, The historical data of the production line is used to extract features using the time series forest algorithm to obtain dynamic time-series features of the production line, including: For any batch Any parameter fingerprint A time series, defined as a length of A sliding window is used to extract a set of statistical features within the window to form a dynamic time-series feature vector of the parameter within a specific time period of the batch. The extracted features include: mean Reflecting parameters Average level within the window; standard deviation Reflecting parameters The degree of fluctuation within the window; the slope of the trend. Obtained through linear fitting, reflecting the parameters The changing trend; approximate entropy Reflecting parameters The complexity and regularity of time series; definition Through this process, the raw data of each batch is transformed into a production line dynamic time series feature dataset composed of dynamic time series features of each section and parameter.

5. The monitoring method for an anti-static foam box production line for a liquid crystal display screen according to claim 4, characterized in that, The dynamic time-series characteristics of the production line are analyzed causally using a decision tree model to generate a mutual intuition rule base for the production line, including: After obtaining the dynamic time-series feature dataset of the production line, the goal is to uncover the causal relationships between the features of different work sections. To this end, a highly interpretable decision tree model is used for analysis. Model training and rule extraction were performed using the dynamic temporal features of all work sections. Using the electrostatic safety performance index of the final product as the prediction target, the decision tree model is trained. After training, the path from the root node to each leaf node is extracted from the decision tree. Each path is a classification rule in the form of IF-THEN. Next, a production line mutual inductance rule base is constructed. These rules are organized, redundancies are removed, and priorities are sorted to form the production line mutual inductance rule base, denoted as . Each rule in this rule base Each contains three core components: triggering conditions, prediction results, and root cause localization. The triggering conditions are constituted by the dynamic temporal characteristics of one or more previous work sections meeting a specific threshold range; the prediction results indicate specific electrostatic safety problems that may occur in future work sections without intervention; and the root cause location clearly identifies the key root cause work section number that triggers this rule. and specific parameter fingerprint number This design directly serves S3, providing targets for generating precise intervention plans.

6. The monitoring method for an antistatic foam box production line for a liquid crystal display screen according to claim 5, characterized in that, The real-time dynamic time-series characteristics of the production line are acquired, and these characteristics are input into the production line mutual inductance rule base for matching to locate the root cause section and parameters, including: Data is collected in real time from sensors at each stage of the antistatic foam box production line and recorded as real-time production line data. The data acquisition source, synchronization mechanism, and preprocessing process are consistent with those in step S1. Subsequently, following the feature extraction method in step S2, dynamic temporal features are calculated on the real-time data stream. To distinguish it from the historical features in S2, the dynamic temporal feature vector extracted from the current real-time data is denoted as... The set of these features is called the real-time dynamic time series feature set. This step transforms continuous real-time signals into structured features that can be used for pattern matching by the rule base. Real-time dynamic time series feature set Input into the production line mutual inductance rule base generated by S2 Real-time matching is performed, which involves searching for all rules. , making Satisfy the rules The triggering condition, once matched successfully, will be determined according to the rules. The root cause location information defined in the code immediately identifies the critical root cause section number that triggers the abnormal trend. and parameters .

7. A monitoring method for an anti-static foam box production line for a liquid crystal display screen according to claim 6, characterized in that, Based on the aforementioned root cause section and parameters, simulated intervention was performed on the shadow production line in the antistatic foam box production line, including: Based on the root cause section located in step S3 and parameters Generate structured intervention plans Its generation principle is to map the abnormal state of the root cause parameter into a specific control command based on the adjustment strategy preset in the historical experience or process knowledge base. A complete intervention program It should include the following core elements: intervention objective, adjustment actions, window of action, and expected verification points: For the intervention target, clearly specify the work section that needs adjustment. and parametric fingerprints ; For adjustment actions, specify the specific operations, directions, and recommended ranges for parameter adjustments; for the action window: define the start time, duration, or end conditions of the intervention; for the expected verification points: specify which subsequent work section and which parameter to monitor to quickly assess the intervention effect; A shadow production line is set up that is completely identical to the main production line in terms of process equipment and control logic. It is used for safety verification of intervention schemes. The main production line is used for actual production, and the shadow production line is used for simulation testing.

8. The monitoring method for an antistatic foam box production line for a liquid crystal display screen according to claim 7, characterized in that, If the simulated intervention effect reaches the expected value, then the main production line of the antistatic foam box production line will be adjusted, including: The generated intervention plan The process involves deploying a shadow production line for simulation verification. The verification process includes two core evaluation dimensions, and quantitative indicators are introduced for judgment, encompassing effectiveness assessment and impact assessment. For effectiveness evaluation, the main focus is on assessing the improvement effect of the intervention program on electrostatic safety and quality, defining the rate of decrease in quality risk indicators. for: ； in, and These respectively represent the implementation of intervention plans on the shadow production line. Before and after, the average value of the key quality risk indicators monitored; when When the validity verification is passed, A validity threshold pre-set based on statistical analysis of historical data; For impact assessment, the main focus is on evaluating the impact of intervention programs on production stability and energy efficiency, defining production cycle fluctuation rate. and the rate of change of unit energy consumption for: ； in, and These represent the implementation of intervention plans on the shadow production line. The average production cycle time obtained before and after; and These represent the implementation of the intervention plan. Before and after the intervention, the average energy consumption per unit of product was statistically obtained; the requirements for the intervention were as follows: and ,in and The preset tolerance threshold is based on production requirements. This is the maximum permissible production cycle time fluctuation, typically set based on the adjustable range of the production plan. It is the maximum permissible rate of change in unit energy consumption, which is usually set based on energy efficiency assessment indicators; An intervention plan is determined only if both of the above assessments are passed. Verification successful; migration to the mainline is permitted.

9. A monitoring method for an antistatic foam box production line for a liquid crystal display screen according to claim 8, characterized in that, Through hierarchical model predictive control (MPC) algorithm-based collaborative optimization, the main production line maintains stable operation under the intervention plan, producing batches of antistatic foam boxes awaiting release, including: Once the intervention plan Or a new optimized solution Once validated on the shadow production line, it needs to be safely and smoothly migrated to the main production line for execution. A hierarchical predictive control architecture is adopted. Specifically, the operation flow of this hierarchical control system includes upper-level controllers and lower-level controllers: The upper-level controller is responsible for strategic rules on a slow timescale; In each optimization cycle, the upper-level controller, based on the overall macro model of the production line and the current state, solves for two key decision variables: the global production cycle time setpoint and the global production takt time setpoint. The first purpose is to coordinate the production rates of each work section and avoid bottlenecks; the second is to allocate energy consumption budgets for each work section. For each work section Set upper limits for energy consumption to achieve energy efficiency management; For the objective function, its optimization objective is to satisfy the intervention plan. Given the requirements, and taking into account both production efficiency and energy consumption, this objective can be formally expressed as a weighted summation function: ； in, This is a cost term characterizing total energy consumption, defined as the estimated total energy consumption of the main production line during the intervention period, calculated using the following formula: , Representing the Estimated energy consumption of each work section during the intervention period; It is a cost term characterizing production capacity, defined as the negative value of total output during the intervention period, i.e. , This represents the expected number of production batches to be completed within this period. Minimizing this cost term is equivalent to maximizing total output; weighting coefficients. and This is pre-set according to the actual production strategy. When production tasks are tight and priority needs to be given to ensuring delivery, the following settings are configured: The optimization objective will tend to maximize output; when the factory is in an energy-saving and consumption-reducing cycle, then the following settings will be implemented: The optimization objective will tend to minimize energy consumption; the output of the upper-level controller is the globally optimal setpoint combination. ; The lower-level controller is responsible for tactical execution on a fast timescale: Each section is equipped with an independent MPC controller, which operates rapidly at a cycle of seconds or less to ensure timely response to process disturbances and setpoint changes. The core task of each section controller is to drive the process parameters of its section to track the set values ​​issued by the upper level quickly and accurately, while maintaining smooth control actions and avoiding impact on the equipment. The optimization problem of the MPC controller at each work section level is solved within each control cycle, and its objective function is typically designed as follows: ； in, For the prediction time domain, it is a preset positive integer representing how far into the future the controller will predict the system output; To control the time domain, it is a preset positive integer that satisfies... This indicates how far in the future the controller will optimize the control increments; It is the output variable vector of the work section; Is the section in the future Predicted values ​​of the output variable vector at time step; It is the corresponding setting value issued by the upper-level MPC; It manipulates the variable vector; Indicates the change in the control quantity; In the future Control increment at any given moment; and The weight matrix penalizes both the output tracking error and drastic changes in the control quantity, and needs to be determined during on-site commissioning based on the dynamic characteristics of each section; index Used to iterate through future moments within the prediction time domain; index , used to traverse the control steps within the control time domain; Through closed-loop interaction between upper and lower level controllers, dynamic collaboration is achieved. When an intervention plan causes a change in the setpoint of a certain section, the upper-level MPC will recalculate the globally optimal cycle time and energy consumption allocation, and issue new instructions to all sections. The lower-level MPC controllers will immediately respond, adjusting their local actuators to ensure that the output of their section tracks the new setpoint. Through this mechanism of upper-level decision-making and lower-level execution, the entire production line can smoothly and stably transition to the new operating conditions after intervention, and continuously maintain a dynamic balance between production cycle time and energy consumption while meeting quality requirements, producing batches of antistatic foam boxes ready for release. .

10. A monitoring method for an antistatic foam box production line for a liquid crystal display screen according to claim 9, characterized in that, Electrostatic safety load verification was performed on the batches of the aforementioned antistatic foam boxes to obtain the electrostatic safety load results, including: From batch Representative samples were selected and placed in a simulated test environment that could simulate specific relative humidity, vibration frequency, and friction conditions. A pulse of a specified voltage was applied using a standardized electrostatic discharge device, and the charge decay time on the surface of the enclosure was immediately measured. and residual voltage ; Measurement results Safety thresholds specified in the product specifications Compare; if satisfied and If the electrostatic safety verification is passed, the batch is deemed to have passed and is permitted to be released.

Images