Film product production intelligent management system based on production whole-process tracking

Abstract

The present application relates to the technical field of production management data processing, in particular to a film product production intelligent management system based on production whole-process tracking. The system comprises an edge collection node, a time delay compensation engine and a dynamic parameter adjustment executor. The edge collection node collects physical quantities and marks a space position stamp based on linear speed. The time delay compensation engine reversely maps downstream defect coordinates to upstream physical space positions according to the transmission path equation between traction roller groups and real-time linear speed, and extracts physical quantity sequences in the corresponding interval. The dynamic parameter adjustment executor inputs the sequences into a process deviation analysis model, takes the Euclidean distance between physical quantities and standard reference curves as a feature, and outputs compensation values to be sent to a controller. The scheme overcomes the data time delay error caused by material transmission distance, realizes the spatial alignment of downstream detection and upstream parameters, converts after-the-fact statistics into in-process intervention, and reduces the continuous waste product generation rate.

Description

Technical Field

[0001] This invention relates to the field of production management data processing technology, specifically to an intelligent management system for thin film product production based on full-process production tracking. Background Technology

[0002] In the field of continuous film manufacturing, existing production management systems typically deploy sensors at key stations such as extruders, casting rolls, and traction rolls to collect physical quantities such as screw speed, casting roll temperature, and traction tension at fixed frequencies. Simultaneously, an optical defect detector is deployed at the winding front end to identify defects on the film surface. When the optical defect detector detects a defect in the film, the existing management system directly downgrades the current winding or triggers a stop command. Throughout the production process, the upstream process parameter acquisition and recording and the downstream defect detection data recording are two independent data streams; the management system only stores these data separately and performs post-production statistical comparisons.

[0003] Because there is a physical transport distance of tens of meters between the film extrusion molding stage and the downstream optical defect detector, the material needs to undergo a considerable amount of time within this range. By the time the optical defect detector outputs the defect coordinates, the film segment that caused the defect has already left the upstream extrusion or stretching station. Existing management systems, limited by the time delay caused by this physical transport distance, cannot reverse the mapping of the downstream detected defect coordinates to the specific physical spatial location of the upstream defect. This results in the inability to extract the extruder screw speed, casting roll temperature, and traction tension sequence that actually corresponds to the defect, leading to a disconnect between defect tracing and closed-loop control. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent management system for thin film product production based on full-process production tracking, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A smart management system for thin film product production based on full-process production tracking, including edge acquisition nodes, a time delay compensation engine, and a dynamic parameter tuning actuator;

[0007] The edge acquisition node synchronously acquires the current extruder screw speed, casting roll temperature and traction tension at a fixed sampling frequency, and marks the acquired physical quantities with a spatial position stamp calculated based on the production line line speed;

[0008] The time delay compensation engine receives the two-dimensional coordinates of the defects output by the downstream optical defect detector. Based on the transmission path equation of the film between the traction rollers and the real-time linear velocity, the two-dimensional coordinates of the defects are reverse-mapped to the physical spatial position of the upstream extrusion or stretching stage. The extruder screw speed, casting roller temperature and traction tension sequence within the spatial position stamp interval of the corresponding physical spatial position are extracted.

[0009] The dynamic parameter adjustment actuator inputs the extracted multidimensional physical quantity sequence into the pre-constructed process deviation analysis model. The process deviation analysis model uses the Euclidean distance between each physical quantity and the standard reference curve as input features, outputs the temperature compensation value and tension adjustment value for the next production cycle, and sends the temperature compensation value and tension adjustment value to the extruder temperature controller and tension controller.

[0010] Preferably, when the edge acquisition node acquires physical quantities, it calculates the real-time movement distance of the film by reading the encoder pulse signal, and sums the real-time movement distance with the pre-stored reference coordinates of the traction roller group to obtain the spatial position stamp;

[0011] During the generation of the spatial location stamp, when the edge acquisition node detects an extruder screen changing signal or a sudden change in traction tension signal, it adds a status flag bit to the corresponding spatial location stamp. The status flag bit is used as the boundary identification basis for removing abnormal data segments during subsequent physical quantity sequence extraction.

[0012] Preferably, when performing reverse mapping, the delay compensation engine obtains the topology parameters of the traction roller group, including the center coordinates, wrap angle and roller surface friction coefficient of each traction roller, and constructs the transmission path equation based on the topology parameters.

[0013] The delay compensation engine converts the horizontal coordinates in the two-dimensional coordinates of the defect into corresponding arc length increments based on the change in the wrap angle, subtracts the sliding compensation length calculated based on the linear velocity and friction coefficient from the vertical coordinates, and calculates the spatial location stamp of the upstream physical space position by combining the arc length increment and the sliding compensation length.

[0014] Preferably, when the delay compensation engine extracts the sequence of physical quantities within the spatial location stamp interval of the corresponding physical location, it takes the calculated spatial location stamp of the upstream physical location as the center point and extends a preset buffer distance to both sides to form the spatial location stamp interval.

[0015] Within the spatial location stamp interval, the time delay compensation engine reorders the collected extruder screw speed, casting roll temperature and traction tension in chronological order, removes redundant physical quantities with inverted timestamps after reordering, and generates a multidimensional physical quantity sequence that strictly increases according to the time series.

[0016] Preferably, the standard reference curve in the process deviation analysis model is generated by applying a moving average filter to the physical quantity sequence of the historical good product production stage. The standard reference curve includes a screw speed reference curve, a casting roll temperature reference curve, and a traction tension reference curve that are consistent with the time length of the multidimensional physical quantity sequence.

[0017] When calculating the Euclidean distance, the process deviation analysis model calculates the difference between the physical quantity at each time point in the multidimensional physical quantity sequence and the physical quantity on the reference curve at the corresponding time point. The difference is squared, summed, and then squared to generate the Euclidean distance of screw speed, Euclidean distance of casting roll temperature, and Euclidean distance of traction tension.

[0018] Preferably, when the dynamic parameter adjustment actuator outputs the temperature compensation value and the tension adjustment value, it inputs the Euclidean distance of the screw speed, the Euclidean distance of the casting roll temperature and the Euclidean distance of the traction tension into a preset multiple linear regression equation. The coefficient matrix of the multiple linear regression equation is dynamically configured according to the process parameter sensitivity matrix corresponding to different film grades.

[0019] The dynamic parameter adjustment actuator generates the temperature compensation value and tension adjustment value by weighted superposition calculation based on the output result of the multiple linear regression equation, combined with the current temperature setpoint of the extruder temperature controller and the current tension setpoint of the tension controller.

[0020] Preferably, when the edge acquisition node attaches a status flag bit, it records the trigger time of the status flag bit and the corresponding spatial location stamp, and distinguishes the type of the status flag bit into a network switching transient flag and a tension anomaly flag;

[0021] When the spatial location stamp interval extracted by the delay compensation engine contains a network switching transient marker, the delay compensation engine uses the spatial location stamp corresponding to the network switching transient marker as a dividing point to divide the spatial location stamp interval into a pre-network switching sub-interval and a post-network switching sub-interval, and only sends the physical quantity sequence in the post-network switching sub-interval to the dynamic parameter tuning executor.

[0022] Preferably, when constructing the transmission path equation, the time delay compensation engine corrects the topology parameters based on the real-time operating status of the traction roller group, wherein the real-time operating status includes the current speed difference and surface temperature of the traction rollers;

[0023] When the current speed difference of the traction roller exceeds the preset speed difference threshold, the time delay compensation engine recalculates the actual wrap angle of the film on the corresponding traction roller surface based on the current speed difference, replaces the wrap angle in the original topology parameters with the recalculated actual wrap angle, generates a corrected transmission path equation, and performs reverse mapping based on the corrected transmission path equation.

[0024] Preferably, during the generation of the standard baseline curve, when the cumulative data volume of the physical quantity sequence of the historical good product production stage reaches the preset data window length, the process deviation analysis model triggers the baseline curve update process.

[0025] During the baseline curve update process, the process deviation analysis model adds the latest acquired physical quantity sequence to the end of the data window, while removing early physical quantity sequences of the same length from the beginning of the data window. The updated physical quantity sequence in the data window is then reprocessed with a moving average filter, and the reprocessed curve replaces the original standard baseline curve.

[0026] Preferably, when dynamically configuring the process parameter sensitivity matrix, the dynamic parameter tuning actuator obtains the resin melt index and target thickness parameters corresponding to the currently produced film grade, and retrieves multiple historical process records in the pre-built sensitivity database that are closest to the resin melt index and target thickness parameters;

[0027] The dynamic parameter adjustment actuator extracts the ratio of extruder temperature adjustment to screw speed change, and the ratio of tension adjustment to traction speed change from multiple historical process records. The extracted ratios are normalized and filled into the coefficient matrix of the multiple linear regression equation.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. This invention assigns spatial location stamps based on production line linear velocity to physical quantities collected by edge acquisition nodes. Combined with the film's transmission path equation between traction rollers and real-time linear velocity, it reverse-maps the two-dimensional coordinates of defects output by the downstream optical defect detector to their upstream physical spatial locations. By extracting the physical quantity sequence within the corresponding spatial location stamp interval and inputting it into the process deviation analysis model, and outputting temperature compensation and tension adjustment values ​​based on the Euclidean distance between the physical quantities and the standard reference curve, it sends these values ​​to the controller. This overcomes the data delay error caused by the physical distance of material transmission, achieving strict alignment between downstream detection data and upstream process parameters in physical space. It transforms post-event defect statistics into in-process dynamic process intervention, reducing the continuous scrap rate caused by process fluctuations.

[0030] 2. By attaching a status marker bit to the spatial location stamp and using it as the boundary identification basis for eliminating abnormal data segments, and combining the wrap angle and friction coefficient of the traction roller group to calculate the sliding compensation length and construct the transmission path equation, the buffer distance is extended to both sides when extracting the physical quantity sequence and redundant physical quantities with inverted timestamps are eliminated. Based on the physical quantity sequence of the historical good product production stage, a standard reference curve is generated after moving average filtering. This eliminates the interference of non-process fluctuation factors such as screen changing transients, corrects the positional offset caused by the film sliding on the traction roller surface, ensures the strict progression of the multidimensional physical quantity sequence in the time dimension, and improves the accuracy of the input data of the process deviation analysis model and the reliability of the system's output compensation parameters under complex working conditions. Attached Figure Description

[0031] Figure 1 This is a flowchart of the edge acquisition node spatial location stamp generation and status marking process of the present invention;

[0032] Figure 2 This is a flowchart illustrating the reverse mapping process for the delay compensation engine defects of the present invention.

[0033] Figure 3 This is a flowchart of the status flag triggering and abnormal data removal process of the present invention;

[0034] Figure 4 This is a flowchart illustrating the construction and dynamic correction of the transmission path equation in this invention.

[0035] Figure 5 This is a flowchart illustrating the standard baseline curve generation and rolling update process of the present invention.

[0036] Figure 6 This is a flowchart illustrating the process deviation analysis and dynamic parameter tuning process of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] Please refer to Figure 1 This embodiment provides an intelligent management system for film product production based on full-process production tracking. It runs on a production line that includes an extrusion molding unit, a casting and shaping unit, multiple sets of traction rollers, an optical inspection unit, and a winding unit arranged in sequence. The system includes edge acquisition nodes, a time delay compensation engine, and a dynamic parameter adjustment actuator. Each functional unit achieves synchronous data interaction through industrial real-time Ethernet.

[0039] In one embodiment, edge acquisition nodes synchronously acquire the current extruder screw speed, casting roller temperature, and traction tension at a fixed sampling frequency, and stamp the acquired physical quantities with a spatial location stamp calculated based on the production line speed. Preferably, the fixed sampling frequency is triggered by a global synchronization clock signal to ensure that the acquisition times of edge acquisition nodes deployed at different workstations are strictly aligned, eliminating time offset errors in multi-source physical quantity data. Specifically, edge acquisition nodes are deployed in the extruder drive unit, the casting roller temperature control unit, and the traction roller tension detection unit, respectively. Each node receives a hard trigger signal from the same clock source and completes the sampling and analog-to-digital conversion of the corresponding physical quantity at the same time, generating an original data frame containing the sampling time, physical quantity value, and corresponding workstation identifier.

[0040] Furthermore, the edge acquisition nodes calculate the spatial location stamp of the corresponding film segment based on the real-time linear velocity of the production line, thus binding the time-dimensional sampling data one-to-one with the physical spatial location of the film. The real-time linear velocity is calculated using the following formula:

[0041]

[0042] in, for Real-time line speed of the production line. The reference roller diameter for the main traction roller. for The number of pulses output by the encoder within the time window. This represents the total number of pulses corresponding to each revolution of the encoder. This is the sampling time interval corresponding to the fixed sampling frequency.

[0043] Based on the calculation results of real-time linear velocity, the spatial location stamp is generated using the following formula:

[0044]

[0045] in, for Spatial location stamps corresponding to the physical quantities collected at any given time. The starting values ​​for the pre-stored reference coordinates of the traction roller group are the reference spatial coordinates of the extruder die exit. For the starting time of the corresponding film segment extrusion from the die, the integral term is the time from the extrusion of the corresponding film segment from the die to... The real-time movement distance at any given moment. The edge acquisition node fills the generated spatial location stamp into the specified field of the original data frame, completing the binding of physical quantity data with the physical spatial location of the thin film.

[0046] refer to Figure 2The delay compensation engine receives the two-dimensional coordinates of defects output by the downstream optical defect detector. Based on the film's transmission path equation and real-time linear velocity between the traction rollers, it maps the two-dimensional coordinates of the defects back to their physical spatial positions in the upstream extrusion or stretching stages. It then extracts the extruder screw speed, casting roller temperature, and traction tension sequence within the corresponding spatial position stamp interval. Specifically, the optical defect detector is deployed at the end of the traction roller group and the front of the winding unit. Its output two-dimensional coordinates of defects include horizontal and vertical coordinates. The horizontal coordinate represents the position of the defect along the film width direction, and the vertical coordinate represents the position of the defect along the film's travel direction. The origin of the coordinates is the starting edge of the optical defect detector's detection window.

[0047] The transmission path equation of the pre-stored film in the time delay compensation engine between the traction rollers is constructed based on the fixed topology of the production line, and the total transmission path length is calculated using the following formula:

[0048]

[0049] in, This is the total transmission path length from the extruder die exit to the inspection window of the optical defect detector. For the first The length of the straight section of film between adjacent traction rollers. This represents the total number of line segments. For the first The diameter of each traction roller, For the thin film in the first The wrap angle on each traction roller, in radians. This represents the total number of traction rollers.

[0050] Based on the transmission path equation and real-time linear velocity, the delay compensation engine reverse-maps the two-dimensional coordinates of the defect to its upstream physical location. The corresponding physical location is calculated using the following formula:

[0051]

[0052] in, This refers to the upstream physical location of the defect. These are the reference space coordinates corresponding to the starting edge of the detection window of the optical defect detector. This represents the vertical coordinate in the two-dimensional coordinate system of the defect. This is the slip compensation length of the film on the traction roller surface. The time delay compensation engine calculates this. Based on this, a corresponding spatial location stamp interval is defined. All physical quantity data, including extruder screw speed, casting roll temperature and traction tension, within this interval are extracted from the database uploaded by the edge acquisition nodes to generate the corresponding multidimensional physical quantity sequence.

[0053] refer to Figure 6 The dynamic parameter-adjusting actuator inputs the extracted multidimensional physical quantity sequence into a pre-constructed process deviation analysis model. The model uses the Euclidean distance between each physical quantity and the standard reference curve as input features, outputting the temperature compensation value and tension adjustment value for the next production cycle, and then sending these values ​​to the extruder temperature controller and tension controller. Specifically, the process deviation analysis model pre-stores standard reference curves that match the dimensions of the multidimensional physical quantity sequence. These standard reference curves include the screw speed reference curve, the casting roll temperature reference curve, and the traction tension reference curve. For each extracted multidimensional physical quantity sequence, the process deviation analysis model calculates the Euclidean distance between each physical quantity and its corresponding standard reference curve. The Euclidean distance is calculated using the following formula:

[0054]

[0055] in, This represents the Euclidean distance between the corresponding physical quantity sequences. For the first in a multidimensional physical quantity sequence The physical quantity values ​​at each sampling point The first in the standard reference curve The baseline value of each sampling point This represents the total number of sampling points for the multidimensional physical quantity sequence. The process deviation analysis model calculates and generates the Euclidean distances for screw speed, casting roll temperature, and traction tension, respectively, and inputs these three sets of Euclidean distances as input features into the preset multiple linear regression equation.

[0056] The expression for the multiple linear regression equation is as follows:

[0057]

[0058] in, This is the output temperature compensation value. This is the output tension adjustment value. The Euclidean distance represents the screw rotation speed. The Euclidean distance for the casting roll temperature. The traction tension is the Euclidean distance. and These are the coefficient matrix elements of the multiple linear regression equation. The dynamic parameter adjustment actuator sends the calculated temperature compensation value to the extruder temperature controller and the tension adjustment value to the tension controller, completing the closed-loop adjustment of the production process parameters.

[0059] In this embodiment, Table 1 is inserted to clarify the reference spatial coordinate parameters of each key station in the thin film production line, providing a reference basis for the generation of spatial location stamps and cross-station coordinate alignment.

[0060] Table 1. Reference Spatial Coordinate Parameters of Key Workstations in the Thin Film Production Line

[0061]

[0062] The reference spatial coordinates in Table 1 take the extruder die outlet as the origin and the film-moving direction as the positive direction. The relevant parameters are pre-stored in the local storage unit of the edge acquisition node and the parameter storage unit of the delay compensation engine, and are used for the initial value calculation of the spatial location stamp, cross-station spatial coordinate alignment and the reference positioning of defect reverse mapping.

[0063] In this embodiment, the spatial location stamp enables one-to-one binding between upstream process parameters and thin film physical sections. The transmission path equation completes the reverse mapping from downstream defect coordinates to upstream production stations, realizing the linkage between defect tracing and process closed-loop control. This overcomes the data flow disconnection problem caused by the physical distance of thin film transmission and completes the full-process operation of the system's core functions.

[0064] In another embodiment, reference Figure 3 When acquiring physical quantities, the edge acquisition node calculates the real-time movement distance of the film by reading encoder pulse signals. It then sums this real-time movement distance with the pre-stored reference coordinates of the traction roller group to obtain a spatial position stamp. Specifically, the encoder is coaxially mounted with the driven end of the main traction roller, outputting pulse signals corresponding to the rotation angle of the traction roller in real time. The edge acquisition node reads the cumulative value of the pulse signals in each sampling cycle, calculates the real-time movement distance of the film within the corresponding sampling cycle based on the number of pulses and the diameter of the traction roller, and sums the real-time movement distance with the reference coordinates of the corresponding station to generate a spatial position stamp corresponding to that sampling moment. This ensures that the calculated spatial position stamp strictly corresponds to the actual film movement distance.

[0065] During the generation of spatial location stamps, when the edge acquisition node detects an extruder screen change signal or a sudden change in traction tension, it adds a status flag bit to the corresponding spatial location stamp. This status flag bit serves as the boundary identification basis for removing abnormal data segments during subsequent physical quantity sequence extraction. Specifically, the edge acquisition node monitors the screen change trigger signal of the extruder control system and the real-time data from the traction tension detection unit in real time via industrial Ethernet. When it receives a screen change trigger signal, it immediately adds the corresponding status flag bit to the spatial location stamp generated in the current sampling period. For traction tension data, the edge acquisition node identifies tension change signals based on preset judgment conditions. The trigger judgment for tension change signals is completed using the following formula:

[0066]

[0067] in, This represents the traction tension value at the current sampling time. This represents the traction tension value at the previous sampling time. This is a preset tension mutation threshold. When the above inequality holds, the edge acquisition node determines that a tension mutation has occurred at the current sampling time and adds a status marker bit to the corresponding spatial location.

[0068] When attaching status flags, edge acquisition nodes record the trigger time and corresponding spatial location stamp of the status flags, and distinguish between network switching transient flags and tension anomaly flags. Specifically, the status flags are stored in reserved fields of the physical quantity data frame, containing three core fields: flag type code, trigger timestamp, and corresponding spatial location stamp. The type code for the network switching transient flag is 0x01, and the type code for the tension anomaly flag is 0x02. Different types of flags correspond to different abnormal data processing rules.

[0069] When the spatial location stamp interval extracted by the latency compensation engine contains a network switching transient marker, the latency compensation engine uses the spatial location stamp corresponding to the network switching transient marker as a dividing point to divide the spatial location stamp interval into a pre-network switching sub-interval and a post-network switching sub-interval. Only the physical quantity sequence within the post-network switching sub-interval is sent to the dynamic parameter tuning executor. Specifically, when extracting the physical quantity sequence, the latency compensation engine first traverses the status flag bit field of all data frames within the spatial location stamp interval. When a network switching transient marker is detected, the spatial location stamp corresponding to the marker is locked, and the original interval is divided into two sub-intervals using this as a dividing point. The pre-network switching sub-interval contains physical quantity data before the network switching action is triggered, and the post-network switching sub-interval contains physical quantity data after the network switching action is completed. The range of the valid sub-interval is determined by the following formula:

[0070]

[0071] in, This is the starting spatial location stamp of the valid sub-interval. This is the termination spatial location stamp of the valid sub-interval. This is the spatial location stamp corresponding to the transient marker of the network switching. This refers to the upstream physical location of the defect. This is the preset buffer distance. The time delay compensation engine only extracts physical quantity data within the valid sub-interval, eliminating abnormal fluctuation data within the sub-interval before screen change, thus avoiding interference from abnormal data caused by transient fluctuations in melt pressure during screen change on the process deviation analysis results.

[0072] In this embodiment, Table 2 is inserted to clarify the triggering rules and corresponding data processing strategies for different types of status flag bits, providing a standardized basis for the boundary identification and removal of abnormal data segments.

[0073] Table 2. Comparison of Status Flag Bit Types and Triggering Rules

[0074]

[0075] The triggering rules in Table 2 are pre-configured in the logic control unit of the edge acquisition node. Different types of status flag bits correspond to different abnormal operating conditions and data processing strategies, ensuring that abnormal data segments can be accurately identified and removed under abnormal operating conditions other than process fluctuations, thus ensuring the purity of the input data of the process deviation analysis model.

[0076] In this embodiment, the data boundary is accurately identified under abnormal operating conditions by using status flag bits. Corresponding rejection rules are formulated for abnormal data caused by non-process fluctuation factors such as screen switching transients and tension changes, so as to avoid the interference of abnormal data on the process parameter adjustment process and improve the stability of the system under complex operating conditions and the reliability of process deviation analysis results.

[0077] In yet another embodiment, reference is made to... Figure 4 When performing reverse mapping, the delay compensation engine obtains the topology parameters of the traction roller group. These parameters include the center coordinates, wrap angle, and surface friction coefficient of each traction roller. Based on these parameters, a transmission path equation is constructed. Specifically, the topology parameters of the traction roller group are defined based on a two-dimensional coordinate system of the production line. The coordinate system has its origin at the center point of the extruder die exit, with the positive Y-axis along the film feeding direction and the positive X-axis along the film width direction. The center coordinates of each traction roller are the absolute coordinate values ​​within this coordinate system. Based on these topology parameters, the delay compensation engine constructs the continuous transmission path parameter equations for the film between the traction roller groups. For the j-th traction roller, the corresponding tangent point coordinates are calculated using the following formula:

[0078]

[0079] in, , Let J be the coordinates of the center of the j-th traction roller. Let j be the diameter of the j-th traction roller. Let be the incident angle of the film entering the j-th traction roller, in radians. The angle of wrap of the film on the j-th traction roller is expressed in radians. , Let be the coordinates of the tangent point where the film enters the j-th traction roller. , Let be the coordinates of the tangent point where the film leaves the j-th traction roller. By sequentially calculating the entry and exit tangent point coordinates of all traction rollers, the time delay compensation engine constructs a continuous transmission path equation covering the entire traction roller group, realizing a full-process digital representation of the film's movement path.

[0080] The delay compensation engine converts the lateral coordinates in the defect's two-dimensional coordinate system into corresponding arc length increments based on the change in wrap angle. It subtracts the sliding compensation length calculated based on linear velocity and friction coefficient from the lateral coordinates. Combining the arc length increments and the sliding compensation length, it calculates the spatial location stamp of the upstream physical location. Specifically, the lateral coordinates in the defect's two-dimensional coordinate system correspond to different positions along the film's width. Due to the wrap angle on the traction roller surface, the film's path length differs at different positions along the width. Therefore, it is necessary to convert the lateral coordinates into corresponding arc length increments. The calculation of the arc length increments is accomplished using the following formula:

[0081]

[0082] in, This represents the arc length increment corresponding to the horizontal coordinate. The horizontal coordinate in the two-dimensional coordinate system of the defect. It is the sum of the wrap angles of the film on all the traction rollers. This represents the total width of the thin film.

[0083] To address the film travel distance deviation caused by the relative sliding of the film on the traction roller surface, the time delay compensation engine calculates the corresponding sliding compensation length. The calculation of the sliding compensation length is accomplished using the following formula:

[0084]

[0085] in, For the sliding compensation length, This refers to the real-time line speed of the production line. This refers to the contact residence time of the film on the corresponding traction roller. The friction coefficient of the traction roller surface is denoted as . The time delay compensation engine subtracts the sliding compensation length from the longitudinal coordinates of the defect, superimposes the arc length increment, and combines it with the transmission path equation to complete the reverse mapping from the two-dimensional coordinates of the defect to the upstream physical space position, thus obtaining the upstream spatial position stamp corresponding to the defect.

[0086] When constructing the transmission path equation, the time delay compensation engine corrects the topology parameters based on the real-time operating status of the traction roller group. The real-time operating status includes the current speed difference and surface temperature of the traction rollers. When the current speed difference of the traction rollers exceeds a preset speed difference threshold, the time delay compensation engine recalculates the actual wrap angle of the film on the corresponding traction roller surface based on the current speed difference. It then replaces the wrap angle in the original topology parameters with the recalculated actual wrap angle, generating a corrected transmission path equation, and performs a reverse mapping based on the corrected transmission path equation. Specifically, the time delay compensation engine obtains the current speed of each traction roller in real time, calculates the speed difference between each traction roller and the main traction roller, and when the speed difference exceeds a preset threshold, it determines that the wrap angle of the film on the surface of that traction roller has changed, corrects the wrap angle parameter, and the corrected actual wrap angle is calculated using the following formula:

[0087]

[0088] in, This is the corrected actual wrap angle. The reference wrap angle in the original topology parameters. Let j be the current linear velocity of the j-th traction roller. The current linear velocity of the main traction roller. The time delay compensation engine reconstructs the transmission path equation using the corrected actual wrap angle, and performs reverse mapping of the defect coordinates based on the corrected transmission path equation, eliminating the impact of wrap angle changes caused by traction roller speed differences on the reverse mapping accuracy.

[0089] When extracting the physical quantity sequence within the spatial location stamp interval of the corresponding physical location, the time delay compensation engine uses the calculated spatial location stamp of the upstream physical location as the center point and extends a preset buffer distance to both sides to form a spatial location stamp interval. Within the spatial location stamp interval, the time delay compensation engine reorders the collected extruder screw speed, casting roll temperature, and traction tension in chronological order, eliminating redundant physical quantities with inverted timestamps after reordering, and generating a multidimensional physical quantity sequence that strictly increases in time sequence. Specifically, the preset buffer distance is set according to the sampling frequency of the production line and the duration of process fluctuations to ensure that the spatial location stamp interval can completely cover all process parameter data corresponding to the defect generation stage. Since data collected from multiple nodes may be out of order during network transmission, the time delay compensation engine reorders all data frames within the interval in ascending order according to the sampling timestamp, traverses the reordered data frames, and eliminates redundant data with timestamps smaller than the previous frame's timestamp, ensuring that the final generated multidimensional physical quantity sequence strictly increases in time sequence and eliminating data timing errors caused by transmission out-of-order transmission.

[0090] Table 3 is inserted in this embodiment to clarify the topological parameters and dynamic correction rules of the traction roller group, providing a standardized parameter basis for the construction and correction of the transmission path equation.

[0091] Table 3. Traction Roller Set Topology Parameters and Correction Rules

[0092]

[0093] The topology parameters in Table 3 are pre-stored in the parameter storage unit of the time delay compensation engine. The correction rules are used to dynamically adjust the relevant parameters of the transmission path equation according to the real-time operating status of the traction roller, so as to ensure that the transmission path equation can accurately match the real-time operating conditions of the production line and improve the position accuracy of defect reverse mapping.

[0094] In this embodiment, the precise digital representation of the film-moving path is achieved through the transmission path parameter equation based on the topology of the traction roller group. By converting the arc length increment and calculating the sliding compensation length, the deviation in film-moving distance caused by the difference in film width position and the sliding of the roller surface is corrected. The wrap angle parameter is dynamically corrected by the difference in the rotational speed of the traction roller, which is adapted to the real-time operating status of the production line. By reordering data and removing redundant data, the temporal consistency of the multidimensional physical quantity sequence is ensured, and the positioning accuracy of defect reverse mapping and the accuracy of physical quantity sequence extraction are improved.

[0095] In yet another embodiment, reference is made to... Figure 5 The standard baseline curve in the process deviation analysis model is generated by applying a moving average filter to the physical quantity sequence of historical good-quality production stages. The standard baseline curve includes a screw speed baseline curve, a casting roll temperature baseline curve, and a traction tension baseline curve, all with a time length consistent with the multidimensional physical quantity sequence. Specifically, the physical quantity sequence of historical good-quality production stages consists of process parameter data corresponding to continuous production stages where defects were not detected by the optical defect detector. The process deviation analysis model applies a moving average filter to this data to eliminate high-frequency noise and generate a smooth standard baseline curve. The moving average filter is calculated using the following formula:

[0096]

[0097] in, This is the reference value for the i-th sampling point in the standard reference curve. The length of the sliding window, expressed in units of sampling points. This represents the physical quantity value at the m-th sampling point during the historical good product production stage. The process deviation analysis model performs moving average filtering on the historical good product data for screw speed, casting roll temperature, and traction tension, generating three standard benchmark curves that are completely consistent with the input multidimensional physical quantity sequence, providing a benchmark basis for the quantitative calculation of process deviations.

[0098] During the generation of the standard baseline curve, when the cumulative data volume of the physical quantity sequence from the historical good product production stage reaches the preset data window length, the process deviation analysis model triggers the baseline curve update process. In this update process, the process deviation analysis model adds the latest collected physical quantity sequence to the end of the data window, while simultaneously removing earlier physical quantity sequences of equal length from the beginning of the data window. The updated data window then undergoes a new moving average filtering process on the physical quantity sequences, and the reprocessed curve replaces the original standard baseline curve. Specifically, the preset data window length is set based on the continuous production time and sampling frequency of the production line to ensure that the data window covers the entire production cycle. The baseline curve update process employs a first-in, first-out (FIFO) rolling update mechanism, and the updated data window is represented by the following formula:

[0099]

[0100] in, For the updated data window, For the original data window, To supplement the length of the latest physical quantity sequence, This is the latest sequence of physical quantities collected during the production of good products. The process deviation analysis model re-executes the moving average filtering process on the updated data window, generates a new standard baseline curve and completes the replacement, realizing the dynamic rolling update of the standard baseline curve, and adapting to the drift problems of working conditions such as screw wear and roller surface aging that occur during long-term operation of the production line.

[0101] When calculating the Euclidean distance, the process deviation analysis model calculates the difference between the physical quantity at each time point in the multidimensional physical quantity sequence and the physical quantity on the corresponding reference curve. The differences are squared, summed, and then squared to generate the Euclidean distances for screw speed, casting roll temperature, and traction tension. The specific calculation process is consistent with the Euclidean distance calculation formula in the aforementioned embodiment. When the dynamic parameter adjustment actuator outputs temperature compensation and tension adjustment values, it inputs the Euclidean distances for screw speed, casting roll temperature, and traction tension into a preset multiple linear regression equation. The coefficient matrix of the multiple linear regression equation is dynamically configured based on the process parameter sensitivity matrix corresponding to different film grades.

[0102] When dynamically configuring the process parameter sensitivity matrix, the dynamic parameter tuning actuator obtains the resin melt index and target thickness parameters corresponding to the currently produced film grade. It then retrieves multiple historical process records from a pre-built sensitivity database that are closest to these parameters. Specifically, the sensitivity database stores historical process records corresponding to different film grades, material parameters, and product specifications. Each historical process record includes a complete process parameter adjustment procedure and corresponding product quality inspection results. The dynamic parameter tuning actuator uses the currently produced resin melt index and target thickness as search criteria to retrieve historical process records from the sensitivity database where the melt index deviation is within a preset range and the target thickness deviation is within a preset range, ensuring the compatibility of the retrieved historical records with the current production conditions.

[0103] The dynamic parameter adjustment actuator extracts the ratios of extruder temperature adjustment to screw speed change, and tension adjustment to traction speed change from multiple historical process records. These extracted ratios are then normalized and added to the coefficient matrix of the multiple linear regression equation. Specifically, the dynamic parameter adjustment actuator performs statistical analysis on all retrieved historical process records, calculating the ratios of extruder temperature adjustment to screw speed change and tension adjustment to traction speed change for each record. The calculated ratios are then normalized using the following formula:

[0104]

[0105] in, These are the normalized coefficient values. This is the original ratio data. This is the minimum value in this set of ratio data. This represents the maximum value among the ratio data. The dynamic parameter tuning actuator fills the normalized coefficient values ​​into the coefficient matrix of the multiple linear regression equation, completing the dynamic configuration of the coefficient matrix and ensuring that the multiple linear regression equation can adapt to the currently produced film grade, material parameters, and product specifications.

[0106] The dynamic parameter adjustment actuator generates temperature compensation and tension adjustment values ​​through weighted superposition calculations based on the output results of the multiple linear regression equation, combined with the current temperature setpoint of the extruder temperature controller and the current tension setpoint of the tension controller. Specifically, the weights of the weighted superposition calculations are dynamically adjusted according to the Euclidean distance of the corresponding physical quantities. The larger the Euclidean distance, the higher the corresponding weight, ensuring that the physical quantity with the greater process deviation corresponds to a larger adjustment range, thus achieving precise closed-loop adjustment of process parameters.

[0107] Table 4 is inserted in this embodiment to clarify the process parameter sensitivity matrix corresponding to different film grades, providing standardized data basis for the dynamic configuration of the coefficient matrix of the multiple linear regression equation.

[0108] Table 4. Comparison of Process Parameter Sensitivity Matrix for Different Thin Film Grades

[0109]

[0110] The sensitivity matrix data in Table 4 is stored in the sensitivity database of the dynamic parameter adjustment actuator. It corresponds to the process adjustment characteristics of different film grades, material parameters and product specifications. It is used to dynamically configure the coefficient matrix of the multiple linear regression equation according to the current production conditions to ensure the adaptability and accuracy of process parameter adjustment.

[0111] In this embodiment, a smooth standard baseline curve is generated through moving average filtering. The standard baseline curve is dynamically updated through a rolling update mechanism, which adapts to the long-term operating condition drift of the production line. Through sensitivity matrix retrieval and normalization based on material parameters and product specifications, the coefficient matrix of the multiple linear regression equation is dynamically configured, which adapts to the production process characteristics of different film grades, improves the accuracy of process parameter adjustment and the system's adaptability to different production scenarios.

Claims

1. A smart management system for thin film product production based on end-to-end production tracking, characterized in that: This includes edge acquisition nodes, a latency compensation engine, and a dynamic parameter tuning executor; The edge acquisition node synchronously acquires the current extruder screw speed, casting roll temperature and traction tension at a fixed sampling frequency, and marks the acquired physical quantities with a spatial position stamp calculated based on the production line line speed; The time delay compensation engine receives the two-dimensional coordinates of the defects output by the downstream optical defect detector. Based on the transmission path equation of the film between the traction rollers and the real-time linear velocity, the two-dimensional coordinates of the defects are reverse-mapped to the physical spatial position of the upstream extrusion or stretching stage. The extruder screw speed, casting roller temperature and traction tension sequence within the spatial position stamp interval of the corresponding physical spatial position are extracted. The dynamic parameter adjustment actuator inputs the extracted multidimensional physical quantity sequence into the pre-constructed process deviation analysis model. The process deviation analysis model uses the Euclidean distance between each physical quantity and the standard reference curve as input features, outputs the temperature compensation value and tension adjustment value for the next production cycle, and sends the temperature compensation value and tension adjustment value to the extruder temperature controller and tension controller. The spatial location stamp is generated using the following formula: in, for Spatial location stamps corresponding to the physical quantities collected at any given time. The starting values ​​for the pre-stored reference coordinates of the traction roller group are the reference spatial coordinates of the extruder die exit. For the starting time of the corresponding film segment extrusion from the die, the integral term is the time from the extrusion of the corresponding film segment from the die to... Real-time movement distance at any given moment.

2. The intelligent management system for thin film product production based on full-process production tracking as described in claim 1, characterized in that, When acquiring physical quantities, the edge acquisition node calculates the real-time movement distance of the film by reading the encoder pulse signal, and sums the real-time movement distance with the pre-stored reference coordinates of the traction roller group to obtain the spatial position stamp. During the generation of the spatial location stamp, when the edge acquisition node detects an extruder screen changing signal or a sudden change in traction tension signal, it adds a status flag bit to the corresponding spatial location stamp. The status flag bit is used as the boundary identification basis for removing abnormal data segments during subsequent physical quantity sequence extraction.

3. The intelligent management system for thin film product production based on full-process tracking as described in claim 1, characterized in that, When performing reverse mapping, the delay compensation engine obtains the topology parameters of the traction roller group, including the center coordinates, wrap angle and roller surface friction coefficient of each traction roller, and constructs the transmission path equation based on the topology parameters. The delay compensation engine converts the horizontal coordinates in the two-dimensional coordinates of the defect into corresponding arc length increments based on the change in the wrap angle, subtracts the sliding compensation length calculated based on the linear velocity and friction coefficient from the vertical coordinates, and calculates the spatial location stamp of the upstream physical space position by combining the arc length increment and the sliding compensation length.

4. The intelligent management system for thin film product production based on full-process production tracking as described in claim 3, characterized in that, When the delay compensation engine extracts the physical quantity sequence within the spatial location stamp interval of the corresponding physical location, it takes the calculated spatial location stamp of the upstream physical location as the center point and extends a preset buffer distance to both sides to form the spatial location stamp interval. Within the spatial location stamp interval, the time delay compensation engine reorders the collected extruder screw speed, casting roll temperature and traction tension in chronological order, removes redundant physical quantities with inverted timestamps after reordering, and generates a multidimensional physical quantity sequence that strictly increases according to the time series.

5. The intelligent management system for thin film product production based on full-process production tracking as described in claim 1, characterized in that, The standard reference curve in the process deviation analysis model is generated by the moving average filtering of the physical quantity sequence of the historical good product production stage. The standard reference curve includes the screw speed reference curve, the casting roll temperature reference curve and the traction tension reference curve, which are consistent with the time length of the multidimensional physical quantity sequence. When calculating the Euclidean distance, the process deviation analysis model calculates the difference between the physical quantity at each time point in the multidimensional physical quantity sequence and the physical quantity on the reference curve at the corresponding time point. The difference is squared, summed, and then squared to generate the Euclidean distance of screw speed, Euclidean distance of casting roll temperature, and Euclidean distance of traction tension.

6. The intelligent management system for thin film product production based on full-process production tracking as described in claim 5, characterized in that, When the dynamic parameter adjustment actuator outputs temperature compensation value and tension adjustment value, it inputs the Euclidean distance of screw speed, Euclidean distance of casting roll temperature and Euclidean distance of traction tension into a preset multiple linear regression equation. The coefficient matrix of the multiple linear regression equation is dynamically configured according to the sensitivity matrix of process parameters corresponding to different film grades. The dynamic parameter adjustment actuator generates the temperature compensation value and tension adjustment value by weighted superposition calculation based on the output result of the multiple linear regression equation, combined with the current temperature setpoint of the extruder temperature controller and the current tension setpoint of the tension controller.

7. The intelligent management system for thin film product production based on full-process tracking as described in claim 2, characterized in that, When the edge acquisition node attaches a status flag, it records the trigger time of the status flag and the corresponding spatial location stamp, and distinguishes the type of the status flag into a network switching transient flag and a tension anomaly flag. When the spatial location stamp interval extracted by the delay compensation engine contains a network switching transient marker, the delay compensation engine uses the spatial location stamp corresponding to the network switching transient marker as a dividing point to divide the spatial location stamp interval into a pre-network switching sub-interval and a post-network switching sub-interval, and only sends the physical quantity sequence in the post-network switching sub-interval to the dynamic parameter tuning executor.

8. The intelligent management system for thin film product production based on full-process tracking as described in claim 3, characterized in that, When constructing the transmission path equation, the time delay compensation engine corrects the topology parameters based on the real-time operating status of the traction roller group. The real-time operating status includes the current speed difference and surface temperature of the traction roller. When the current speed difference of the traction roller exceeds the preset speed difference threshold, the time delay compensation engine recalculates the actual wrap angle of the film on the corresponding traction roller surface based on the current speed difference, replaces the wrap angle in the original topology parameters with the recalculated actual wrap angle, generates a corrected transmission path equation, and performs reverse mapping based on the corrected transmission path equation.

9. The intelligent management system for thin film product production based on full-process tracking as described in claim 5, characterized in that, During the generation of the standard baseline curve, when the cumulative data volume of the physical quantity sequence of the historical good product production stage reaches the preset data window length, the process deviation analysis model triggers the baseline curve update process. During the baseline curve update process, the process deviation analysis model adds the latest acquired physical quantity sequence to the end of the data window, while removing early physical quantity sequences of the same length from the beginning of the data window. The updated physical quantity sequence in the data window is then reprocessed with a moving average filter, and the reprocessed curve replaces the original standard baseline curve.

10. The intelligent management system for thin film product production based on full-process production tracking according to claim 6, characterized in that, When dynamically configuring the process parameter sensitivity matrix, the dynamic parameter tuning actuator obtains the resin melt index and target thickness parameters corresponding to the currently produced film grade, and retrieves multiple historical process records that are closest to the resin melt index and target thickness parameters in the pre-built sensitivity database. The dynamic parameter adjustment actuator extracts the ratio of extruder temperature adjustment to screw speed change, and the ratio of tension adjustment to traction speed change from multiple historical process records. The extracted ratios are normalized and filled into the coefficient matrix of the multiple linear regression equation.

Images